The question of what algorithmic trading really is often gets blurred in heated discussions of the risks and dangers involved. We have already seen several regulatory initiatives aimed at containing the risks of what is referred to as “algorithmic trading,”
and we’re likely to see more, so it has become increasingly important to be clear about scope and context.
The problem is that algorithmic trading is in fact variously defined; often it’s understood to include all automated trading, including the use of decision-making software. Algo “purists” like myself would prefer that the term be restricted to execution
algorithms, i.e., standards such as VWAP, together with the custom algos developed, in most cases, by brokerage firms for their traders and clients. Additionally, smart routers can be considered as multi-market execution algos, and are often used in conjunction
with VWAP or other strategies.
The key characteristic of all such execution algos is that they do not make original trading decisions, but merely determine execution tactics for pre-existing orders of given, limited size. Given this tight constraint, these algos are unlikely to unleash
the kind of market havoc that we’ve seen in the course of several well-publicized ”algorithmic trading incidents” over the past few years.
Today, execution algos play a vital role in the trading activity of almost all institutional asset managers on the electronic markets of North America and Europe, and increasingly in Asia-Pacific. But unfortunately these algos have been swept up in the furor
over the dangers posed to market stability by automated trading engines in general, and high-frequency trading (HFT) techniques in particular. With this latter, much wider category of algos, original trading decisions are made by the software itself, and the
potential for unpredictable and damaging outcomes is far greater.
A defective decision-making algorithm, if presented with input data or operational circumstances for which it has not been adequately tested, may decide to do almost anything, and continue to do it for as long as those responsible for monitoring the algo
allow. We’re all familiar with the consequences of such incidents from the news stories, and they are of course potentially largest in the HFT context: many erroneous orders can be sent at a rate of thousands per second before “low-frequency” human supervisors
notice and respond to the effects.
The seriousness of some of these cases helps explain the regulatory concerns over automated trading, especially HFT. The definition problem then hits us, because the regulators and politicians often loosely use the term “algorithmic trading” to describe
their area of concern: one frequently sees the descriptors “algo” and “HFT” being used almost interchangeably. But, as discussed above, the decision-making algorithms that drive HFT engines bear almost no relationship to the VWAP and implementation shortfall
algos that are used to work fund managers’ large buy and sell orders over time. So before we get heavy-handed regulation applied to everything that’s labeled “algo trading” (and, by the way, how to make such regulation workable is another question that has
yet to be answered clearly), it’s perhaps worth some additional effort to clarify this distinction with the relevant decision makers.