Algorithmic trading is supposed to optimize investment strategies, taking out the subjective influence of a trader making minute by minute decisions, and high frequency trading (HFT) is known to be lightning fast. But people’s preference for round numbers has a startling effect on trade volumes, clumping activity at regular intervals without a rational explanation.
“Recurring periodicity of crowded trading activity increases as the ‘roundness’ of time marks increases. One hour marks are the most attractive for clustered activity, followed by half-hour marks and then by 10-minute and finally, 5 minute marks. This preference of increasing roundness points to the human nature of the observed phenomenon,” write Rutgers School of Business professors John Broussard and Andrei Nikiforov in their recent paper “Human Bias In Algorithmic Trading” (h/t Empiritrage).
Round numbers usually show uncertainty
The round number hypothesis, which simply says that people confronted with uncertain quantities tend to guess round numbers, has been well-documented in finance. You will almost never see an analyst set a price target like $7.13 (as opposed to $7 or $7.5) for the same reason you wouldn’t schedule a meeting at 10:32. Specifying a PT down to the cent or a meeting time down to the minute implies a level of precision that is rarely available.
The latest Robinhood Investors Conference is in the books, and some hedge funds made an appearance at the conference. In a panel on hedge funds moderated by Maverick Capital's Lee Ainslie, Ricky Sandler of Eminence Capital, Gaurav Kapadia of XN and Glen Kacher of Light Street discussed their own hedge funds and various aspects of Read More
A similar effect can be seen in the price of stocks with very low trade volumes, but it normally disappears as trade volume, options, and active shorts improve price discovery. While it isn’t surprising that trade volumes aren’t uniform throughout the day, the level of intraday periodicity measured by either volume or number of trades is striking.
Algorithmic trading: VWAP may cause one trader’s bias to spread
Broussard and Nikiforov investigate an alternative explanation that this behavior allows Algorithmic traders to reduce their trading costs and defend from better informed investors, but they don’t find any evidence for this argument.
But the periodicity doesn’t necessarily mean that every Algorithmic trading is guilty of having a bias for round numbers. Lots of strategies look at the volume weighted average price (VWAP) of a stock when deciding whether or not to make a trade. If just a few major traders have a round number bias, the prevalence of VWAP would cause algorithmic traders without the whole number bias to naturally clump around the same intervals to reduce their market impact.
Even if this periodicity doesn’t seem like a problem, it’s a good example of how human error can be introduced into supposedly objective algorithms.