The Gateway to Algorithmic and Automated Trading

Everything you ever wanted to know about SGX... but were afraid to ask.

Published in Automated Trader Magazine Issue 18 Q3 2010

Recent months have seen Singapore Exchange (SGX) making a raft of announcements relevant to Automated Trader readers. New colocation services, upgraded technology infrastructure, commodity cross listings – you name it and SGX seems to be doing it. Andy Webb, Automated Trader’s Founder caught up recently with Bob Caisley, Head of IT & Chief Information Officer and Rama Pillai, Head of Intermediaries and Market Access SGX, to get a little more detail on this flurry of activity.

Andy Webb: One of the first things our readers always want to know about is what sort of welcome they can expect from an exchange. So let me start by asking you that - what is SGX's policy on automated and algorithmic trading?

Rama Pillai: SGX embraces this community because we feel they add a different dimension and value to the mix of liquidity participants in our marketplace. We always try to support that kind of activity by virtue of our membership structure and application process so as to facilitate that kind of business on the exchange. For example, we offer both remote and direct memberships and we permit both classes of member to operate a broad range of different strategies on our markets. Proximity hosting services have been available in Singapore for some time, but as you know we have recently announced that we will be offering colocation services in the near future to cater for the growing need for low latency access.

Bob Caisley: Our approach to automated trading is reflected in the numbers; at present automated and algorithmic trading typically represent some 25-30% of total volume activity on our derivatives market. The corresponding automated volume on our cash market is considerably lower, but our pending move to Genium INET on the cash market will obviously facilitate new automated trading entrants - as well as encouraging existing automated traders on our derivatives markets to extend their activities onto the cash market as well.

Andy Webb: Automated and algorithmic traders generally place considerable demands upon an exchange's technology infrastructure in terms of message volume etc. Do you anticipate the need for appreciable additional investment in order to accommodate them?

Rama Pillai: Clearly the automated trading community is a demanding one, but we regard this as something that goes with the territory and is something that we will deal with and accept. The needs of automated traders are clearly very specific, so we want to ensure that we are able to deliver access when and how they need and be efficient in our dealings with them.

Furthermore, this is something that we have already geared up for -- it didn't just happen by accident. We are very deliberately moving into this space with the objective of catering to this market and community. We are accustomed to dealing with a diverse and demanding user base. For example, if you look at the overall client demographic for our derivatives market approximately 80% of all our customers come from outside Singapore -- in other words they are highly international in nature.

Bob Caisley: I'd agree in that it certainly creates extra pressure, but it is pressure that we were prepared for as we could see it coming. You only have to look at what has been happening elsewhere in the world over the last decade or so, where there have been significant market structure changes and new exchange membership types evolving. This was most apparent in the USA originally, though similar changes have also occurred in Europe more recently. I would anticipate that process being even more collapsed in terms of timeframe in Asia.

Being able to observe what has happened in the USA and Europe is a significant advantage in that we can learn from mistakes made in other regions and avoid making them ourselves. So while there is undoubtedly pressure on us to deliver the functionality, reliability, speed and capacity required by automated/algorithmic traders, we are at least able to do this in the context of what has happened elsewhere already and learn from the precedents.

Andy Webb: At what point was it apparent to SGX that the trading world was changing and a new response was needed?

Rama Pillai: I think the tipping point was probably around 2004. Until then we still operated in the derivatives markets using open outcry, but our last trading pit closed in 2004. We had of course seen the writing on wall well in advance of that and had started gearing up our technology and getting the right systems in place as well as working with the trading community to assist them in migrating from open outcry to electronic trading. So I would say that that at least a year before the final pit closed we were already working actively with the community on electronic migration and subsequently raising our game to accommodate higher-frequency automated and algorithmic traders.

Andy Webb: Are the SGX cash and derivatives systems completely separate in terms of technology? If so, how does that play out as regards arbitrage?

Bob Caisley

Bob Caisley

Bob Caisley: Our two trading engines are indeed accessed through two separate gateways. However, the derivatives that SGX offers are not based upon our cash market so the arbitrage point you raise doesn't really apply. For example, in the case of our futures contracts we are competing with markets such as Tokyo and Osaka for global traders' volume in international instruments such as Japanese government bond and Nikkei 225 futures.

However, we are looking to build up derivatives based upon our own cash market instruments in the future so the sort of arbitrage you allude to will be possible.

Andy Webb: What operating system does your core matching technology run on?

Bob Caisley: The Click XT matching engine provided by NASDAQ OMX that we currently operate runs on OpenVMS using Intel Itanium chips and can only run on this specific chip. However, the next generation NASDAQ OMX trading engine - Genium INET - that we will be deploying as part of our "Reach" initiative will run on any commodity Intel CPU and for that we will be using Red Hat Enterprise Linux as the operating system in conjunction with Oracle as the database.

Initial testing of the new platform showed impressive results; in May we conducted a benchmark test with our technology partners at HP's Singapore Capacity Planning Centre and established an average order response time of 90µs "door-to-door". That represents the fastest execution capability currently available globally.

Andy Webb: What is the physical configuration of your matching engines?

Bob Caisley: On the cash market we have two partitions and going forward we will have four partitions; all these partitions will run concurrently as matching engines. So for example you might have all the warrants in one partition and half the cash stocks in another and the remaining cash stocks in another. We are currently working out which securities will be allocated to which partition when we move to a four partition set up.

This partition concept is a good way of extracting maximum performance from a matching engine. The benchmark tests we conducted on the new trading engine indicate that we will be able to do 1 million order book changes per second on just one partition, so obviously across four partitions that equates to an ability to handle 4 million changes per second.

Andy Webb: In view of the partitioning scheme, if one is pairs trading is there an appreciable performance difference if the two stocks concerned are on the same partition?

Bob Caisley: If both stocks are in the same partition then the trade will execute as a single transaction (spread order) instantaneously. So if you wanted to buy Singapore Airlines and sell Tiger Airways as one transaction - so that either both or neither happen - then that could be conducted as one transaction.

However, if the stocks are in different partitions the trade cannot be executed as a single transaction; the two legs have to be entered independently, so there is a possibility of legging risk. That is one of the reasons that we devote a lot of effort to minimising the matching engine response time across partitions so that any legging risk is also kept to the absolute minimum.

Andy Webb: What is the SGX's attitude to sponsored access to the exchange?

Rama Pillai: Brokers are accountable for managing the risk of their clients. They can offer sponsored access to their clients, but client trading activity has to be routed via the brokers' pre-execution checks or filters. However a growing number of clients both on the cash and derivatives side (and also brokers wishing to support global business flows) are looking for direct sponsored access without having to go through these pre-execution checks. So we are currently working on a sponsored access project whereby brokers will be able to on-board clients through a risk filter at the exchange level, but risk monitoring and management will remain the responsibility of the brokers.

In a very limited respect we already allow this sort of direct access in our derivatives market for a restricted and particular group of participants. The broker has to be prepared to accept the client risk and also has to be able to demonstrate to us that they have near real-time post trade checks in place. However this process is currently very much the exception rather than the norm.

Bob Caisley: As Rama mentions, an SGX sponsored access project is already underway and a technology vendor to support this has been selected. At present we are working on the delivery timeframe, but we expect to have this up and running within twelve months, if not sooner.

The risk management rules for this process will run at the gateway level on the server that accepts the incoming transaction from the customer. The appropriate clearer will set the parameters on that server for the broker to ensure that position limits are not exceeded. There will be certain triggers based upon the maximum level and if certain percentages of that maximum level are exceeded alerts will be tripped and/or orders may start being throttled.

However, while the primary responsibility for managing these gateways will rest with the clearer, the exchange is obviously not going to abrogate all responsibility for the process. Therefore we will also have controls of our own running on the gateway that will deal with preventing problems such as fat finger errors.

Andy Webb: The new colocation service you will be providing - is this something SGX will be offering directly or only via third parties? And how will it operate?

Rama Pillai: SGX will be offering colocation itself and clients will subscribe to the service directly with the exchange, but we will use partner vendors to help deliver the service.

Bob Caisley: The colocation service will run from the same room in which the exchange's matching engines reside. In order to ensure a level playing field, we will be using fixed length cables so it won't matter precisely where a participant's servers are located in the room, they will still be the same wire length from the matching engine as everybody else. Therefore whether a trading server is 1m or 50m away from the matching engine the latency will be identical.

Rama Pillai: While the key attribute of the colocation service is obviously low latency, when we roll it out we also want to build a community of participants. So in our colocation concept we are trying to build a facility that will cater for most (if not all) of our market participants who will want to be in this facility whether they are motivated purely by minimising latency or because they wish to participate in a variety of tradable products.

An important consideration is that something like 50% of SGX's revenues are derived from products or services originating on other markets. As a result, a lot of participants are not just looking to trade on SGX - they also want to arbitrage against other products or markets in Asia.

The colocation service will facilitate these international arbitrageurs by providing access to regional trading via cross connection services. For example, SGX has a venture with Chi-X to offer a dark pool in Asia called Chi-E that will be available to participants from within the colocation facility.

Andy Webb: What about traders wishing to use your new colocation facility to access other exchanges in the region?

Rama Pillai: We recognise that many of our participants are trading multiple markets - especially if you look at our derivative product suite where most of our products are based outside Singapore. For example, we have stock index futures from Japan, Taiwan, India and we are building out into China. Therefore our participants want to be able to arbitrage into other liquidity pools and/or they wish to trade cash markets elsewhere.

Therefore if we want to build a community we have to recognise that they will want to participate on other markets as well. As a result, to the extent it is feasible and practical, we are very open to having different marketplaces or liquidity pools within our own colocation site. In addition to Chi-E, we are also working on the cash equity trading side to build an order routing network within ASEAN. To date, four exchanges have decided to participate - Malaysia, Thailand, the Philippines and ourselves. This will be facilitated by an ASEAN gateway that will route orders out to the respective markets and that will be available from our colocation site.

Therefore a broker connecting to us can also cross connect to these other exchanges. In addition we have announced that we are looking to put overseas points of presence in place in various key locations. This will facilitate order flow to our markets from places such as London, New York, Chicago and Tokyo. These communication hubs can work both ways; so for instance we can facilitate arbitrage trading from Singapore into Japan. The key from our perspective is providing the tools for traders to be able to improve the efficiency of their trading across a portfolio of activities.

Bob Caisley: We will be putting a point of presence in a data centre in Japan where a lot of traders are active, so the intention is to encourage those traders to start using our trading solutions here in Singapore. In addition, some brokers have advised us that there is also a significant tax benefit to the reverse flow. They tell us that while a server is physically based in Japan every transaction going through it is subject to a Japanese securities tax, but a server based in Singapore is not. As a result, traders with servers in our colocation centre in Singapore could be transacting on our trading engine but also using our link to take orders into our Japanese data centre and from there (for example) directly into the Osaka Securities Exchange without being liable for Japanese tax.

Andy Webb: What has been the take up to date of the SGX colocation service?

Rama Pillai: The service is only due to go live in the first quarter of next year and already 90% of our first phase has been sold out with 41 participants having signed up, of which 21 are either brokers or distributors looking to host on behalf of clients.

Bob Caisley: In response to this we have started work on preparations for our phase 2 colocation facility to cover ongoing demand. We are in the fortunate position of designing these colocation facilities from scratch so that has given us an excellent opportunity to really think through future proofing issues and hopefully avoid becoming victims of our own success. For instance, we are able to plan optimal and equal cable lengths now in order to accommodate greater physical distances between later phases of the facility and our matching engine. That's extremely important in view of the problems some other exchanges have experienced in maintaining a level playing field in a low latency environment with colocation.

Andy Webb: How much will the colocation service cost?

Rama Pillai: We are offering colocation as a two tier service with the lowest latency racks that will give you the fastest market access costing SGD12, 000 per month for a standard 42U rack. We will also be offering another category of racks for those not needing the lowest possible latency that will cost SGD6000 per rack per month. These prices are for the rack alone, with additional charges for electricity (as standard 4.5kva is delivered to each rack), networking etc.

Andy Webb: How do you deal with the issue of technology redundancy?

Bob Caisley: SGX obviously has disaster recovery facilities consisting of a primary data centre and a standby one. All critical services run hot/hot, so if there is a failure in one trading engine it automatically fails over to the other. The two trading engines are connected by our dark fibre backbone so participants would still be able to trade if the primary trading engine had an outage.

However should the whole primary data centre suffer an outage then those collocated there would be unable to trade unless they were also co located elsewhere for redundancy and could use their back up lines into our standby data centre. (Though our own backup site would function as usual and trading as a whole would continue.)

An interesting corollary to this is whether SGX should also offer a redundant colocation service domiciled in the same facility as the standby site containing our back up matching engine. From a historical perspective, it would appear that high frequency traders might not find this particularly useful. We have observed that in the rare event of even minor trading disruption this category of market participant tends to be very conservative as regards operational risk management in that they usually suspend their activities immediately and don't opt to resume trading for a while. As a result, they certainly haven't been clamouring for a back up colocation facility alongside our standby matching engine.

Andy Webb: What networking equipment does SGX use? Are you primarily a single vendor operation?

Bob Caisley: We are mostly a Cisco shop, but our choice of hardware is based upon performance, price point and customer delivery. Therefore for our colocation facilities we are very seriously considering the use of Voltaire's InfiniBand switches as they are the fastest currently available. However, for our backbone and corporate switches and the rest of our network we use Cisco throughout.

An interesting innovation here is the launch of Cisco's new Nexus 5000 series range of switches, which are intended to deliver ultralow latency. These largely consist of layer two switching and we'll be introducing them for general network applications.

By contrast the Voltaire switches are layer three and have been a key element in delivering the 90µs response time of our new trading engine that I mentioned earlier. Ultimately we are driven by performance and price and so if Cisco do not innovate sufficiently at layer three to compete with this we'll look elsewhere. This isn't optional for us as we have to go for maximum performance in terms of trading because that is what our customers demand.

Green innovation at SGX

As we went to press, SGX announced that K-Green Trust has commenced trading on its Mainboard, under the ticker symbol "K-Green". Lawrence Wong, Executive Vice President and Head of Listings at SGX, told us: "We welcome K-Green Trust to SGX. This is also the first Green Infrastructure Trust listed on our exchange. It underlines our attractiveness as a listing venue for the fast growing sector of infrastructure assets in Asia."

K-Green Trust is a business trust established and sponsored by Keppel Integrated Engineering (part of the Keppel Group). At listing, K-Green Trust will own three assets, namely Senoko Waste to-Energy Plant, Tuas Waste-to-Energy Plant, and Ulu Pandan NEWater Plant. Thomas Pang, CEO of Keppel Infrastructure Fund Management and Trustee-Manager of K-Green Trust, tells us: "With growing awareness of environmental issues worldwide, there is greater demand for 'green' infrastructure assets. K-Green Trust is well placed to capture these opportunities in the global drive towards sustainable development, and thus represents an opportunity to participate directly in this growing market segment."

K-Green Trust will not be raising any funds in its listing. To meet the shareholding spread requirement, around 51% of the K-Green Trust will be distributed to Keppel Corporation (KCL) shareholders, by way of a dividend in specie based on one KGreen Trust unit for every five KCL shares. K-Green Trust is expected to have an estimated market capitalisation of S$730 million. This will bring the total number of Infrastructure Business Trusts on SGX to three, with a combined market capitalisation of close to S$1.5 billion. It also brings the total number of listings to-date on SGX to 762.

Andy Webb: How do you handle support and real-time maintenance? Is this done in-house or outsourced?

Bob Caisley: We have a relationship with each of our key vendors - such as Cisco for networking and EMC2 for storage. However, we have also recently signed a major five-year IT infrastructure outsourcing agreement worth SGD110mn with HCL Technologies. Under the agreement, HCL will provide SGX with infrastructure support and management services including our "Reach" initiative. HCL will be our interface to Cisco (HCL are Cisco accredited) and other vendors. Therefore, going forward the first line of support for us will be HCL, but we will obviously also have priority support from Cisco as well as other key vendors.

Andy Webb: How do handle forward capacity planning and how straightforward is upgrading or expanding your infrastructure capacity?

Bob Caisley: We always aim to have an absolute minimum of 100% spare capacity over historical peak loading, but at present we are running at more like 400-500% spare capacity. Upgrading core production systems is never a ten minute job because given the mission critical nature of these systems means it should always be approached with suitable caution. Therefore, while we've changed an Itanium processor overnight before because we had to, we'd normally expect a major server change to take a weekend or so. A key factor here is availability of expertise; it is not easy to find OpenVMS and Itanium specialists compared to finding people who understand Red Hat Linux and commodity Intel technology.

Andy Webb: Has that expertise issue been significant in determining your change of matching engine technology?

It's certainly been a factor. I also suspect that NASDAQ OMX who produce the current platform think it's a strong factor. I assume that they have experienced pushback when attempting to sell it - particularly in regions where there is no or only limited OpenVMS expertise. Hence, I suspect the logic behind the migration of their next generation of technology to commodity Intel CPUs that will run more mainstream operating systems.

Andy Webb: What do you offer in terms of historic data?

Rama Pillai: On the derivatives side we provide time and sales historical data going back at least 10 years. On the cash market side we have some of that but we also want to add on additional stuff like corporate news and actions that have an impact on price. That is something we're trying to build up a database for, to see if we can overlay it with actual price data.

Bob Caisley: A lot of companies now collect their own depth of book history by subscribing directly to exchange feeds. Or they work with key data vendors. A new client coming on would have to go and buy the data elsewhere. However this is a service that we have been thinking about internally as we have seen the other exchanges doing this.

We have the data anyway because we have to keep it as a legal obligation. Such as being able to prove the priority or sequence of events around in order. We have to keep "all financially relevant data so it could even be stuff miles away from the traded price in the book".

Andy Webb: What's your prognosis for the next year? What do you see as your largest challenges and how will you handle them?

Rama Pillai: Many algorithmic traders from the US and Europe have enjoyed almost unfettered market access. It is only recently that sponsored access has become more prominent whereby they have to go through filters either at the broker or exchange level. Many of these participants are now grappling with the task of how they are going to deal with the change. How can they adapt their programming, trading models and business models to make sure that they can remain profitable?

From our perspective we are keen that any risk filters introduced will be as transparent as possible, so that their businesses can continue to operate properly without jeopardising their bottom line. It is clearly not in our or their interests for us to add unnecessary layers of latency, so one of the challenges we face will be delivering the right balance in terms of products, technology and services that will allow them to join our markets. A further challenge or opportunity for us is that many market participants tend to make Singapore their first (or even only) Asian stopping off point, so their expectations of SGX are typically high.

I also believe we will see a further increase in high frequency trading activity on SGX. As mentioned earlier, this is already established on our derivatives market and I can only see this activity increasing. Furthermore, the new cash market technology infrastructure will almost inevitably also attract more high-frequency trading as well.

Bob Caisley: I think frequency trading is an area where the rest of the world changed towards deregulation, whereas exchanges in Asia did not - so we do not yet have that type of naked access. There are regulatory pressures that will increase around this, but we believe we already have the necessary checks and measures in place to handle the situation. We are fortunate in having a regulator that has a very global viewpoint, but I suspect we may see other trading centres having to conform with the sort of regulatory position that we see in Singapore today.

Andy Webb: Thank you both very much for your time.

Rama Pillai