HFT is not inherently evil, in the guns-don’t-kill-people-people-kill-people sense. Good evidence for that is in a new white paper available on High Frequency Trading (HFT) from the sovereign wealth fund department of the Norwegian central bank, NBIM. I think it is very well written and an excellent guide to the issues of HFT.
HFT is inevitable, in a technical sense. Faster and cheaper computers and communications necessarily disturb old systems of trading which had relied on a human market maker. Sadly, all of the old problems of the market, such as kiting stocks,and pump-and-dump market manipulation have been recreated in the HFT universe, without the equivalent regulatory tools to catch and prosecute the owners of criminal algorithms.
My proposal is that all HFT algorithms be given to the market regulators to be stress tested in a sandbox environment before being allowed to trade in the markets. The SEC can run the algortihm against others in current market conditions, and in adverse liquidity conditions, to see what the reaction and interactions will be. After the SEC signs off, that version of the algorithm is trusted to enter the market.
Would it be better to push the burden of testing back to the algorithm owner? Yes, there is some minimum amount of testing that must be expected before the algorithm is even given to the SEC for approval. The SEC should be sharing use cases, scenarios and trade simulation data with the public. However, only the SEC is going to have access to all current and future algorithms at the same time. As with drug interactions, it could be that two algorithms which are benign market participants in isolation start to play of each other in a toxic way when active together.
In the sandbox, activity such as order stuffing, momentum ignition and other criminal acts should stand out clearly, as should other attempts to game the system. These algorithms aren’t going to get out into the wild, and probably shouldn’t even get submitted. The stress tests on market activity – flash crashes, liquidity volatility, etc. should catch other algorithms that would be liquidity sinks in problem times, rather than functioning in a market making capacity.
There are problems with the proposal, not least of which is that it will require the construction of a multi-market sandbox and coordination across regulators. It will also require regulators to step up to an operational cycle of nightly builds and regression tests, automated code signing, etc. on their part. That’s OK, regulators have to keep up or fail. Nobody likes a failed regulator.
And there will be an arms race of algorithms trying to detect the sandbox, play nice while inside, and then become more aggressive in the wild. This is also inevitable, and will require handing over the source code as well as the executable to the SEC for analysis and escrow.
Trust and transparency should be table stakes to participating in the captial markets. We’ve built up a large infrastructure to make sure that trustworthy people are financial advisors, work as investment managers, etc. and we also have tried to put internal controls on trading activity. We have reached the point where internal testing is no longer sufficient to maintain trust.