Why have regulators mooted stress testing for algorithms? Haven't they got enough on their plates?
The US ”flash crash' on 6 May 2010 highlighted to many just how fragile equity market structure has become and the systemic risk that algorithmic trading can pose.
In the aftermath of the crash – which saw the Dow Jones Industrial Average index plummet by almost 1,000 points in a matter of minutes before rebounding just as fast – many market participants pointed the finger at high-frequency traders and differences in exchange protocols for dealing with big market swings.
But a joint report from US regulators the Securities and Exchange Commission (SEC) and the Commodity Futures Trading Commission (CFTC) lay the blame at a mutual fund, which sparked the pandemonium with a single algorithmic order.
The execution algorithm was programmed to sell 75,000 E-Mini S&P 500 futures market but without regard to time or price. High volatility at the time the order was placed – caused by market uncertainty over the possibility of Greek government debt default – led the algo to sell 35,000 E-Mini contracts in less than 14 minutes.
The shock that this order caused on the futures market quickly filtered through to the equities market, where a number of brokers temporarily withdrew from the market and some market makers widened their spreads, reduced available liquidity or exited the market entirely.
The fact that a pretty standard algo order from a well-established money manager could trigger such mayhem, forced regulators and other market participants to acknowledge that safeguarding the equity markets required a broader approach than imposing restrictions on high-frequency traders.
So what has actually been proposed with regards to testing of algorithms?
In the US, there have been no proposals for regulators to test algos directly, but other proposals by the CFTC and SEC would ensure an algorithmic order is fit to hit the market.
In February 2011, a report by the joint SEC-CFTC advisory committee on emerging regulatory issues made 14 recommendations relating to the regulatory response to the events of the flash crash. One of these was to urge the CFTC to impose strict requirements on trading firms that use algos themselves or support clients' algorithmic trading activity.
According to the report, such requirements would “ensure that algorithmic firms…demonstrate that there had been a careful evaluation of how an algorithm would operate in a number of scenarios that engender high market volatility”.
Just a few weeks later, the CFTC proposed that trading firms should provide the details of price collars, throttles and a ”kill button' that are built into their derivatives trading systems to limit the risks associated with DMA. CFTC commissioner Scott O'Malia noted that the proposals – which were out for comment until the end of March – would “serve as the foundation to any future proposed rules concerning testing and supervision requirements related to algorithmic trading”.
Most recently, during a speech at industry body SIFMA's Compliance and Legal Society Annual Seminar, SEC chairman Mary Schapiro said that automation review policies (ARPs) – a set of guidelines introduced after another market crash in 1987 to ensure market participants have adequate oversight of their systems – could be formalised and made obligatory. ARPs currently take the form of policy statements, and are therefore not binding.
Mandating ARP requirements would require trading venues, clearing houses, securities depositories and information processors to acquire and test systems that are necessary to run their operations, and may enforce an extra layer of risk checks for algorithms at the trading venue level.
“Today, with risks including algorithm-generated volume surges and malevolent hackers still very much with us, I believe the SEC should consider making ARP compliance mandatory,” said Schapiro.
Is there any appetite from regulators in other markets to impose mandatory stress testing of algorithms?
In its MiFID review consultation process, the European Commission put forward a proposal that would require firms engaged in automated trading to notify authorities of the algorithms they use, which would include giving details on its design, purposes and functioning.
While the proposals are vague, it seems regulators would not have the ability to intervene by making changes to an algo's functionality; they would only be ”notified' of their function. But if this proposal were to be adopted in the EC's final draft of MiFID II rules (expected at some point this summer), stress testing of algos could be the logical next step.
“Stress-testing of venues and the algos that access them is likely to become the norm; as the search for liquidity becomes more time sensitive, our market structures and reporting standards for regulatory oversight needs to be robust,” Kay Swinburne MEP for Wales and member of the European Parliament's Economic and Monetary Affairs Committee, told The TRADE in April.
Meanwhile, Australia's financial regulator, the Australian Securities and Investment Commission, recently made proposals in its own equity market structure consultation that would require market participants to ensure that algorithms that they or their clients use are appropriately tested before use, document the logic used and have appropriate testing procedures.
Click here to vote in this month's poll.