The recent flash crash in the gold market, first reported by ValueWalk on January 6, is now being attributed to an intentional high frequency trading (HFT) algorithm and not a “fat finger” mistake, according to Eric Hunsader, founder of market software firm Nanex LLC. In this flash event the price of gold dropped over $30 in one second, a rare move indeed given the history of the gold market. Hunsader estimates the value of the trades in question at $500 million.
Mr. Hunsader’s key insight is the fact that the trading algorithm paused during the $30 move and then continued its selling. In other words, it wasn’t a fat finger hitting the sell button once, but rather the finger hitting the sell button many multiples of times. Technology companies such as Microsoft Corporation (NASDAQ:MSFT) and Apple Inc. (NASDAQ:AAPL) have long assumed that a mistake can easily happen with one click, but a double click is an indication of intention. Mr. Hunsader is identifying what he believes to be a well-intentioned HFT algorithm not only due to the pause in trading, but also in the numeric behind the trade grouping.
HFT algorithm breakdown
In a report he posted to the Nanex website (figure 1), Mr. Hunsader breaks down the algorithm into a millisecond chart and groups similar trades from the same assumed trader into a 338-trade grouping. While the trades appear to the naked eye to be from a random set of traders due to the outward appearance of a randomization of the trade sizes (211, 186, 120, 193, 97, 193, 137, 112 and 109 to be precise). However, the pattern repeats itself, the sign of a well-organized trade algorithm. The grouping Mr. Hunsader identifies is an apparent method the HFT algorithm uses that obfuscates the singular identity of the trader by creating a seemingly random sequence of trades. But the algorithm isn’t random; it is a pattern in disguise, is Hunsader’s thesis.
Mathematical algorithms, unlike humans, trade using identifiable mathematical patterns. In other words, a human can execute a series of randomized trades without a discernable pattern. If an algorithm has a repeating mechanism of some sort it often follows a pattern in terms of trade size.
Another interesting element that Hunsader researched is his observation that the HFT algorithm was aware of the exchange mechanisms to prevent crashes and came right to the line, but didn’t cross the line that would have triggered a market halt. “What is disturbing about this algorithm, is that it carefully waited so as not to trip the CME Group Inc (NASDAQ:CME)’s stop logic and halt (trading),” Hunsader wrote. “The halt was from the more lenient volatility circuit breaker after the price declined $30 in less than a second. This algo appears to have been more concerned about preventing an immediate halt, rather than getting the best prices.”
New York markets arbitrage strategy for gold
What’s also interesting in Hunsader’s work is his comparison to other markets. He notes that New York markets were quicker to react than the Chicago futures markets. Market makers typically operate a spread / arbitrage strategy. When they buy gold on one market they may look to immediately sell a related product on another market to hedge their risk and lock in the spread differential between the bid and ask as profit. The market making software can be programmed to look at different markets to determine risk management and pricing variables. Thus, if the gold market is moving lower and the silver market remains in place, this would represent an arbitrage opportunity to buy gold and sell silver. What Hunsader’s research indicates is the HFT algorithm in question might have been more narrow in its nature.
1. February 2014 Gold (GC) Futures
2. February 2014 Gold (GC) Futures (same chart as above, shorter time frame)
3. February 2014 Gold (GC) Futures (same chart as above, shorter time frame)
4. March 2014 Silver (SI) Futures
5. SLV ETF trades
6. GLD ETF trades