Illegal HFT trades

CODEX Entry 2451: Illegal HFT trades


Quote stuffing

The May 6th 2010 flash crash brought unwanted attention to the possibility of ‘latency on demand’ being offered, or created, by large HFT traders. The delay is the time between quotes reaching the HFTs’ servers via direct feeds, versus when the quotes reach the Security Information Processor (SIP). The SIP is the Consolidated Tape that most investors pay to access in order to trade quotes. This is not the same as traders taking advantage of latency in the infrastructure or markets, it is artificially generated latency, which is illegal. The delay between when the public markets received their quotes and when the big traders secured theirs, through colocation and high bandwidth feeds, ballooned out to twenty-six seconds during the peak of the crash. Effectively, certain traders algorithms could assess the market for nearly half a minute, an eon in trading, before the rest knew what was happening.

This is achieved by ‘quote stuffing’ between the exchange and the SIP. The line, in 2010, could take up to 10,000 quotes per second. HFTs place thousands of matching orders through two or more books to clog up the line, without taking any net positions. Once the number of quotes exceeds 10,000, the system starts to create buffers, effectively queuing quotes.

The SEC’s report presented the crash as a rare anomaly where a type of resonance was created between certain algorithms, specifically triggered by a large E-mini order from Waddell & Reed. While this order did not create the crash, the on selling of these orders by other HFTs created a cascade effect. These HFTs rapidly offloaded these E-minis without the algorithmic stop gaps that Waddell & Reed had in place. In terms of quote stuffing only one individual trader, a lone autistic trader, Navinder Singh Sarao in Hounslow, London was ever prosecuted for this trillion dollar dip in the markets. The SEC report also only presented the crash in periods of one second intervals. However, Nanex, which collects and sells 22 billion market data points daily, was able to break the trades down into smaller segments to present a more disturbing picture.

They were able to show that a trial for the flash crash had been done a week earlier at 11 a.m on April 28th 2010. They then discovered, and continued to report numerous latencies in individual stocks, lasting around 100 miliseconds throughout 2011-2014. These were labelled mini-flash crashes. Mini-flash crashes involve individual stocks going down and back up 5% – 20% for between 1 and 5 seconds. As computer speeds have increased, these gaps have grown even shorter. Within these one second V cycles, at the millisecond level, there are half a dozen micro-V cycles. So, it is possible to orchestrate a large return within that single second on the shares, by buying each micro dip and selling each micro peak. Stocks can be swiftly selected when they have strong alphas to the index. This means that they generally move in line with a broader index. As the HFT has the data on the rest of the index they can know, with almost complete certainty, the coming movement in the delayed stock. The stocks are also selected for their thin bid to offer book, making it relatively inexpensive to move the price in that moment. In this manner it is possible, in one second, to double your money on a stock that barely moves for the other 86,000 seconds of the day.

Traders can now place orders in under 60 millionths of a second. On individual stocks the large players are sometimes being given 100,000 millionths of a second (0.1 seconds) lead on quotes ahead of the public markets. This is an epoch at these time scales Funds can design, scale, complete, and then hide transactions within 5 milliseconds. Goldman Sachs’ state of the art data centre sits in an unmarked windowless building at 1300 Federal Boulevard, Carteret, New Jersey. The same building that houses the Nasdaq data servers. This colocation ensures no latency.


Data leaks and frontrunning

Light travels from Washington to New York in 2,000 millionths of a second. It takes an algorithm 400,000 millionths of a second to read a press release, prepare a strategy and implement it. Yet new trading based on the Fed’s press releases are being tracked less than 2 milliseconds after their official release. The value of receiving this information early works in the same way as delaying the market’s access to quotes. A chance to frontrun the market. If the interest rates will go up 25 BPS, bonds can be dumped before the market is aware. Despite numerous procedures put in place by the Fed to ensure a fair release of their press releases, some traders are clearly getting early access to the information, and are front running the market.