Best execution in Asia: is it only about the algorithm?

Algorithms must be adapted for the Asian markets. But this alone is not a guarantee of superior performance, writes Kirsti Suutari, global business manager, Algorithmic Trading at Reuters.

Asian markets are flourishing in relation to the US and Europe, drawing interest and investment from all parts of the world in the perpetual search for alpha. Market performance has been worthy of the attention – growth in some Asian markets has exceeded 140%, returns over 2007 are expected to range from 18- 22% and the IPO market is sizzling. The impact on trading volumes has been astounding, with markets such as Hong Kong, Singapore and Malaysia seeing a fourfold increase in turnover over three years, while others doubled in the same period.

The involvement of internationally experienced investors and the increase in market volumes has meant that trading models that have succeeded in other markets have been imported into Asia. But we know that the Asian markets differ from North American and European markets in many ways. The regulatory environment is different, spreads are wider and there are special auctions, queuing, intraday breaks, different order types, and varying volume patterns. Traders therefore recognise that their automated trading models cannot be imported from US or European markets without adaptation, even for liquid markets like Japan and Australia.

It’s about the model
So, if you’re using an adapted algorithm and best execution eludes you, do you blame the algorithm? Examining the model is a good first place to look if you have suspect executions. Making sure you are using the right algorithm for the issue, market and environment is an essential starting point. But even then, the model itself may not be optimised.

The most effective algorithms are those that have been most thoroughly researched and most stringently tested. This involves a few ingredients to complement the necessary quantitative expertise:

• Research data – complete and accurate market-specific historical information is essential to develop the most effective models. Inaccurate or incomplete data will create an ineffective or even flawed model. Research data must be comprehensive to be valuable and must include all the inputs that have an effect on prices. These include the tick history data, as well as corporate action and reference data, economic indicators and news, preferably all synchronised around the same time-stamping system to preserve the temporal links between cause and effect.
• Testing tools – comprehensive testing such as back testing, realistic situation testing and Monte Carlo simulations can minimise performance anomalies prior to release of a trading model. All require realistic, appropriate data and testing tools. The last testing ground for an algorithm – and not the only one – is in the live market.

It’s not just about the model
Even when your algorithm has been well researched and tested, executions can be missed. In these instances, the decisions that the algorithm is making may be missing opportunities because of speed or efficiency issues. Several factors can affect the performance of automated trading systems, excluding latency incurred at the exchanges (which is assumed to be the same for all):

• Communications – distance to market may be too great, bandwidth inadequate, or networks inefficient.
• Processing – CPU speeds, devices and memory, operating system overhead, or application design may be sub-optimal.
• Component performance – the system performs only as well as its slowest contributing element, such as the data feed or analytics calculations.
• Design and integration efficiency – four top medal contenders in a relay will still lose the race if they drop the baton.

All these factors can be addressed. An action list might read as follows: upgrade technology and deploy the latest versions of software as a first obvious step; switch to direct feeds to minimise data latency and enable quicker decisions; examine your data and technology for compatibility and consolidate around a single data model and symbol system to eliminate translation; simplify your design to reduce risk and complexity and ensure your system components are running like that winning relay team; measure and monitor your system performance to identify and control evolving latency; locate systems close to the exchange to reduce transport latency; and don’t ignore capacity metrics, especially in this world of escalating data update rates. The equation is simple: the quicker and more agile your systems, the better the performance of your trading strategies. This explains why so much time and investment is now going into making systems faster. The ‘arms race’ we have seen in Western equities markets is no less relevant in Asian markets, and may be even more so since competition for the available opportunities is intensifying.

The whole is greater than the sum of the parts
Asian markets are in a state of transition, with an evolving regulatory environment, increasing uptake of technology like algorithmic trading systems and DMA, and new liquidity pools such as crossing networks and ‘dark pools’. Best execution is achieved within the available market parameters – rather than in spite of them – and as these change the algorithms must keep pace, as must the systems that drive them, or opportunities and best execution will be lost.