Asia and Latin America: the new frontier for algo trading

Asia and Latin America: the new frontier for algo trading

The silent growth of algorithmic trading has been one of the most powerful trends in securities and derivatives markets in recent years. But while plenty of attention has been paid to this in North America and Europe, it is a lesser known phenomenon in the smaller markets of Asia and the emerging markets.

Algorithmic trading requires a combination of liquid markets, technical infrastructure and institutional support that are only patchily available in these regions. Yet exchanges and brokers in Asia and Latin America are eager to attract this kind of firm, because of the high trading volumes they can generate.

Many algo traders, of course, rely on high frequency trading – myriad trades in a fraction of a second to take advantage of tiny pricing advantages or inefficiencies in the market.

Such firms have made considerable inroads in Asia in recent years. “Volume is driven by a combination of local and global players, with the latter group dominant,” says Rama Pillai, head of intermediaries at Singapore Exchange. “At the same time, a number of global players are locating themselves in Singapore or other parts of Asia, so the line between local and global participants is rather blurred.”

High frequency traders accounted for 26% of SGX’s derivatives trading in the first quarter of this year, against 15% for the same period a year ago.

Asian gateway

One attraction of SGX is that several of its contracts can be used as proxies for markets elsewhere – in India or Japan, for example – that are more difficult to enter, either because of regulatory or technology obstacles. The exchange wants to become the gateway to Asia for algorithmic traders.

“We want to offer access not just to Singapore products, or Japanese products, or Indian products, but to a wide range of products in the derivatives, securities and commodities space across Asia. Our products cover 80% of Asia,” says Pillai.

Many of today’s high frequency and algo trading firms have grown out of options market-making firms in Chicago and Holland.

Some have had great success, but entering the attractive Asian environment – with its wide spreads and high involvement by (typically less sophisticated) retail investors – has not been without its casualties.

Citadel Investment Group, the Chicago high frequency specialist hedge fund, closed its Tokyo office in December 2008 after heavy losses in the financial crisis, cutting 12 staff in Tokyo and 25 in Hong Kong, an office it kept open.

Global Electronic Trading Co (Getco), a leader in high frequency trading in the US and Europe, has taken a more measured approach, gradually expanding its presence in Asia, particularly last year, to ensure familiarity.

The technical battlefront

It is not just for their high volumes that Asian exchanges seek out algorithmic trading firms. For these companies, predicting the market is not the only battleground. They wage a constant war to reduce latency – the time taken to execute trades. This makes them demanding customers, but lucrative ones – they will purchase additional facilities such as colocation and enhanced order access.   

Tokyo leads on matching speed

Over the last year Asian exchanges have worked hard to solve the infrastructure issues preventing algo traders from becoming more widely involved.

Much attention has focussed on the January launch by the Tokyo Stock Exchange of its new Arrowhead platform, developed by Fujitsu, which, it is claimed, will enable participants to execute trades in only 5 milliseconds.

In February Merrill Lynch unveiled Japan’s first third party colocated service for high frequency trading and direct market access to multiple execution venues.

High frequency flow is already crucial to Japan’s exchanges. Despite significant delays in matching in the past, a UBS report last year estimated that about 30% of Japanese equity trading is now high frequency. Levels for derivatives are comparable.

The Osaka Securities Exchange will move its futures and options trading engine to Tokyo in 2011, meaning that firms will get lower latencies when trading its derivatives.

High profile upgrades to matching engines have taken place at the Tokyo Stock Exchange and the Singapore Exchange. The Australian Securities Exchange has recently announced a new platform for high frequency firms.

Figures measuring latency – the time between a trade being placed by a trading programme and it being filled at the exchange – are notoriously difficult to interpret. Latency can be introduced by the network, the software, or the computer processor at the matching engine. Where it is measured, typically by exchanges and technology providers wishing to market their services, it is very rare for traders to achieve those speeds in practice.

But here are figures given by some Asian exchanges, along with the speeds which were common until recently.

Tokyo Stock Exchange
before January 2010: 3 seconds
after Arrowhead: 5 milliseconds (ms)

Osaka Securities Exchange
60 milliseconds

Singapore Exchange
until recently: around 200ms
now: around 15 ms

Australian Securities Exchange
now: around 300ms
target for 2010-11: 0.05ms

Hong Kong Exchanges and Clearing
now: around 15ms

The problem for the exchanges is satisfying the technical needs of the customers. Compared with their Western counterparts, Asian exchanges have severe limits on order throughput – the quantity of orders a single participant can process each second before flooding the exchange’s matching engine.

“The throughputs on exchanges in the region can range from a mere five to approximately 50 orders per second,” says Alan Donoghue, CEO for Asia of Nyenburgh in Singapore. The Dutch day trading and arbitrage trading company has been active in the region for three years.

Exchanges that rush to provide the frills of high frequency trading without speeding up their own matching engines are therefore putting the cart before the horse.

One example of such a frill is colocation, where exchanges allow traders to place their servers alongside the exchange’s matching engine to reduce the distance (and therefore time) that messages must travel.

But as long as throughput to the matching engine is so limited, the millisecond increases in trading speed offered by colocation are ineffective.

A second obstacle concerns the speed with which traders can receive market data from the exchanges to feed into their algorithms, informing the trades. The amount of data the exchanges can publish makes a huge difference to algorithmic traders and, once again, is a function of the exchanges’ infrastructure.

“If a firm needs 100 meg to run a trading line (for a massive wall of pricing data) and this is virtually impossible to instal for its office in Asia, it can’t trade in the way it does in Chicago or New York,” says one former prop trader.

Fear of unfairness

A third problem is regulatory. “On top of these physical constraints there are the regulatory constraints that require you to have a ‘responsible person or designated trader’ for each session,” says Donoghue. If firms want to expand their throughput by using multiple sessions, this means they cannot.

“And most exchanges in the region have harsher member requirements, costs and certainly take longer in application than their US and European counterparts,” Donoghue adds. Remote membership initiatives have done little to solve this problem.

In part, these regulatory obstacles come from the exchanges’ wariness about being seen to give high frequency or algorithmic traders an advantage.

“Regulation is one of the most inhibiting factors: regulators naturally have questions about opening themselves to sophisticated foreign participants whose market participation may require exchanges to change the fundamental principle of not allowing preferential treatment amongst participants,” says Chris Morris, director of Aequitas Associates, a London-based consultancy to exchanges.

Asian derivative markets like South Korea, Hong Kong, and mainland China are very retail-orientated. Investors there fear that fast-working black box models put them at a disadvantage; regulators are reluctant to risk retaliation from this lucrative segment by pushing services like colocation.   

Competition creeps in

“One of the reasons that facilities for high frequency trading are less developed in Asia is that brokers and traders have little leverage in lobbying the exchanges,” explains Christopher Morris of trading consultants Aequitas Associates in London.

There are very few alternative execution venues in Asia where participants can threaten to take their business, so power is concentrated in the hands of the national exchanges. But alternative trading venues are beginning to become established.

In August 2009, Europe’s leading alternative trading platform for equities, Chi-X, announced a partnership with Singapore’s SGX to launch Chi-East, the first so-called ‘dark pool’ in the Asia Pacific region. In March Chi-East secured a licence to operate in Australia.

Hong Kong Exchanges and Clearing, for example, was loth to comment for this article on the services it offers high frequency firms.

Vendors pave the way

In such a disparate operating and regulatory environment, the role of technology firms as a way into the region is crucial. SGX has licensed 12 software vendors to help get traders up and running quickly.

“We get advanced orders from clients who want to trade in Hong Kong at ‘any price’,” explains Benjamin Bécar, SunGard’s algo trading specialist in Hong Kong. “But this order type is not authorised by the exchange. It can be confusing therefore for, say, a Japanese investor, who is used to sending this kind of order. That’s why they come to us.”

Technology firms like SunGard allow trading models to interface with the vagaries of the local operating environment, such as tick size, lot size, opening and closing auctions and market order types – as well as questions as simple as whether there is a lunch break, and what type of trading can be done during the break.

Explains Bécar: “It takes time and resources for vendors or investors to create or adapt algorithms for a specific market. Some harmony, like that provided by a united trading association, would help the industry grow a lot.”

Brazil finds the right mixture

Latin America’s derivatives market contrasts with Asia’s in being highly concentrated. Since the merger in 2008 of Brazil’s futures exchange, BM&F, with its stockmarket and equity options exchange, Bovespa, this one venue has hosted over 85% of Latin America’s derivatives volume, dwarfing even the next-biggest markets, in Mexico and Argentina.

And BM&F Bovespa is on a roll. Helped by an unprecedented wave of interest rate speculation fuelled by the impending resignation of central bank governor Henrique Meirelles, the exchange blasted away its own records in March, with an average 4.17m contracts traded a day in its equity options segment and 3.1m in its futures division.

BM&F Bovespa is now the world’s fifth biggest derivatives exchange group, with 159m contracts traded in March. Part of this rapid growth is down to algorithmic and high frequency traders, which the exchange has been wooing since early 2009, as part of a wider effort to grow foreign participation in derivatives flow from the current 20% to nearer the 35% it enjoys in equities.

Drawn by the triple attractions of liquidity, simpler regulation and high volatility, firms have been coming to Brazil.

DMA takes off

Direct market access (DMA) – where brokers provide traders with direct access to an exchange, via a unique user ID – has surged since its launch in 2008.

By March this year, 12.8% of average daily derivatives volume, or 779,000 contracts a day, was from DMA, up from 3% or 82,000 in January 2009, according to the exchange. The vast majority of this is from algo funds, though not all of those are high frequency traders. In March high frequency firms accounted for 3.5% of total trading, 207,000 contracts a day.

Link Investimentos, an independent Brazilian brokerage, estimates that trading by high frequency traders grew over tenfold between February and October 2009, helped by latencies that dropped from 85 milliseconds last March to around 10-15 milliseconds by year end.

Brokers are arriving in increasing numbers. In November UBS became one of the first global investment banks to offer customised algorithmic trading for international clients, initially in equities. It has now been joined by JP Morgan.

They join established domestic providers Interfloat and Link Investimentos (which UBS is rumoured to be close to acquiring).

Several foreign traders are active – accounting for around 70% of high frequency volume – including leaders like Getco and Jump Trading.

“Algorithmic traders employ a variety of strategies on these contracts: arbitraging the main and mini indices against the underlying contracts,” explains Mateus Bertti, head of IT modelling at BM&F Bovespa. “Spread strategies are popular – arbitraging one maturity against the other – as are trading the Bovespa index portfolio on the spot market against its futures.”

Bertti expects future growth to spread to interest rate futures.

North-South superhighway

A big impetus for the uptake of DMA was the mutual order routing agreement with CME Group, introduced either side of new year 2009. The deal created a low latency intercontinental pipeline spanning the 5,000 miles from Chicago to São Paulo. US brokers could connect to BM&F Bovespa via the CME Globex platform, while Brazilian firms could trade CME products through BM&F’s electronic system GTS.

Once traders were confident that the trading environment was reliable and there was money to be made, the introduction of colocation in June 2009 was a second important development.

“Now these participants are happy the market is liquid enough and that they can make profits, they’re moving to us to colocate,” says Bertti. “Message volume has increased hugely, driven by these foreign players.”

Meanwhile, the exchange has connected with four software vendors, enabling algorithmic traders to have their orders routed directly. SunGard won a licence to provide this so-called ‘Model 2’ access in March, joining emerging markets specialist Marco Polo Networks, Bloomberg Tradebook and the Brazilian firm Cedro Market & Finances.

“The introduction of DMA Model 2 access in particular was important, in that buy side traders no longer need to rely on broker infrastructure for access to the BM&F exchange order book,” says Laurence Latimer, managing director for trading and client connectivity, Americas at SunGard in New York.

Some investors like this because it means their orders do not have to pass through the infrastructure of potential competitors, which might be able to observe their sensitive trading activity.

“Hedge funds can be anywhere – São Paulo, New York, London – sending algorithmic orders. They just need a local member to sponsor them on the exchange,” Latimer explains.

Regulatory support

Additional support for high frequency trading has come from Brazil’s clear regulatory framework. The regulator has been straightforward from the beginning about how foreign firms can participate. All foreign access is broker-sponsored, allowing a client to use the broker’s exchange ID, but requiring orders to pass through at least one level of risk management.

In DMA Model 2 access, the broker provides this risk management screening. For funds that use colocation under Model 4 access, the risk management is provided by the exchange. This prevents trading being disrupted by thousands of erroneous orders placed by a high frequency firm.

For BM&F Bovespa to become established as a fully mature hub of algorithmic and high frequency trading, there is still considerable progress to be made. But with the basic building blocks in place, this looks likely to be rapid.