Type to search

Trading technology Articles

Will My Next Trader Sit In A Data Centre?

Assessing the potential for trade automation on the buy-side desk. By Mark Northwood, Global Head of Equity Trading at Fidelity Worldwide Investment.
Mark Northwood 14When FIX launched in 1998, buy-side desks at traditional asset managers were still adding direct phone lines to their dealer boards, and order and execution management systems (OEMS’s) were often spreadsheets, copied from paper order tickets with actual time stamps. The order routing process of the day was this: [buy-side reads order –> calls/copies details -> sell side re-enters order].
FIX enabled a rapid evolution, starting with the removal of the fallible transmission of order details between humans. The next evolutionary step saw the buy-side trader given the choice to remove the sell-side trader from the process by routing an order with FIX to a box in a data centre where an algorithm (algo) took charge of order placement decisions. The next step could be to remove the buy-side trader from the process as well, resulting in a “no touch” flow.
We know that automation has revolutionised many industries including our own with exchange order matching, market making and arbitrage activity all now handled by machines. So could it work on a centralised trading desk servicing many investment disciplines, and is this where the pursuit of “best execution” is leading us?
The buy-side trading process
Regulation sets us all the goal of achieving “best execution”. This term seems valid for small marketable orders, but acquires extra dimensions when applied to large institutional orders. At Fidelity Worldwide Investment we target an optimal balance between net price, volume, speed and certainty for each order. Simple enough, right? Of course it isn’t, as the optimal result is impossible to determine in the complex, competitive system known as the market. An order book in the market is generally in a state of “fragile” equilibrium: a stable, mean reverting quote distribution persists until one (or more) trader’s algo is too aggressive, triggering a volatility spike as other algos react instantly. Such action by one individual’s algo may be in their interest but can disrupt the market for others and tempt them to dial up their aggression as well, adding fuel to the HFT machine. “Best execution” means continuously assessing how the strategy is behaving, and adapting it.
When an asset manager trades for its clients, the best result combines: 1) a good decisions to buy and sell positions, and 2) intelligent and cost-efficient implementation of those decisions.
The two steps have been segregated into specialist skills as firms have grown. Measuring the impact of different types of trade activity on fund performance is essential. This work is most effective if it spans the entire process, assessing the trading decision itself and not just the execution step.
Traders can then use that analysis to identify ways to improve the decision making step. For example that analysis might show that the so-called fast money tends to over-price news announcements as it anticipates the response by investors, and price reversion is likely hours or days later. Traders could then use such evidence to recommend a disciplined approach when reacting to the “market moving stories” being pushed out by the media and brokers to encourage impulsive activity. So the starting point for further automation is to accurately profile each decision from the Portfolio Managers (PMs) in terms of the many internal attributes which may be relevant to it, plus the external conditions prevailing at the time it is received. This profile will drive the selection of a target execution strategy, incorporating “fair price” bands linked to expected alpha. This would provide more value than an estimated cost, and gets us closer to what could be termed “best execution”.
Human versus machine
Today the trader’s desktop is dominated by large screens. New products and services are pushing more and more information through a graphical interface to a human user with only two eyes, expected to pick what is important from millions of pixels, with a reaction time of 200 milliseconds. The human trader has effectively been at full capacity for years; a bottleneck at the centre of an enormous flow of information, which needs to be tied together, understood and acted upon. That is not to say that human involvement is no longer vital. It is, and I will show this by comparing the relevant capabilities of a good trader to a concept Automated Trader, or “AT”.
The AT envisaged here is more than an auto-router sending small %ADV orders to an IS algo*. The AT would comply with handling practices for common order combinations, determine an execution strategy from the order profile, continuously check the current state of each order against its target state and assess possible alternative states. Such a “cognitive” OEMS is not yet a reality for asset managers, but many of the required components are in active use at firms engaged in HFT and other systematic trading desks. The proposition is that the time has come to explore whether AT could play a significant role on the buy-side. A subjective comparison of the two contenders follows.
Broking is a human specialty, so could an AT interact as effectively with a broker, or would a good buy-side trader have better access to information about client flows because of their relationship? A solitary AT in the high and low touch world of today would struggle. But if other firms embraced the concept and more market practices were dragged into the electronic age, AT could operate in a world of message traffic between venues offering a variety of continuous and auction mechanisms for different types of transaction. Block deals could perhaps be offered directly to investors by institutions in specially structured dark auctions, following targeted marketing based on investors’ recent search history… Back in the real world, much of the time the other side of the trade doesn’t yet exist. Sometimes a deal will only come together with a little selling, and that remains something that certain humans have always been good at!
The optimal process
An engineer would solve this by designing a process with the human trader doing what he or she does best, and the AT playing to its strengths: The traders monitoring the filtered stream of highly processed information on a single dashboard highlighting exceptions, opportunities arising from others’ clumsy trading , and illiquid names. The AT seamlessly blending new orders into the Auto-OEMS, maintaining integrity of the allocation process without incurring delay and cost from intraday bookings, and making continual adjustments to the price and aggression settings on its trading algorithms as conditions change.
The target operating state for the AT is derived from the analytical profiling described earlier, as are the breaches which trigger intervention. The notable risks would be mitigated through multiple safeguards: constraining the interactions between human trader and machine, independently supervising every part of the process in real time, and instantly suspending activity which violates the normal state.
A significant re-engineering of the process would be required, but it is already best practice to record and track the key human inputs, so decisions become electronic instructions flowing through a system. As each component of an AT was developed, it would be inserted into the process under the watchful eye of the specialist trader. The obvious impediment is the huge investment in time and money which would be required, as this would be supplementary to the time and cost of running the business. Experience of the complexity of handling institutional orders, and with getting today’s algos to perform consistently suggests this effort cannot be underestimated. Regulators, for obvious reasons, would also have concerns and are likely to add restrictions and a burden of proof before such trading technology could be widely adopted by the buy-side. Car drivers make fatal mistakes every day, but there will be public outcry the first time an autonomous vehicle causes a crash.
The buy-side trader will still have a key role to play in years to come, as will the OEMS but the partnership will evolve as each learns new skills from the other. Over time, the trading system will learn what it needs to take over the order handling and micro-decisions during order execution. Traders will increasingly focus on supervision of the system performance, and using much improved data to consult with PMs, advising on the trade timing and strategy, steering complex orders, assessing market activity for opportunities and only rarely intervening directly in an errant trade execution.
*jargon: algorithmic trading strategy targeting an execution price close to the market price when the order is placed, minimising the Implementation Shortfall.
More Buy-side Interviews Here


You Might also Like