HFT businesses can now easily predict stock movement for lower-priced equities
The authors of a recent article from Princeton University claim that high-frequency trading (HFT) returns and durations are predictably “huge, systemic, and widespread.” Let’s see how it can help algorithmic trading companies.
According to researchers Yacine At-Sahalia, Jianqing Fan, Lirong Xue, and Yifeng Zhou in their working paper titled “How and when are high-frequency stock returns predictable”, high-frequency trading firms can forecast the direction of a trade, over the following five seconds, with 63 percent accuracy.
The 58-page study also reveals that lower-priced companies that are “less liquid, less volatile, and less connected with the aggregate market” have more predictable returns and trade directions.
Researchers have discovered that some times of the day are more predictable than others, with the end of the trading day being one of these occasions.
The researchers note that spreads usually narrow throughout the day and that volatility is often higher at the start and finish of trading days and lower in the middle. The trading day is divided into three sessions: an opening session from 9.30 am to 10 am, a midday session from 10 am to 3 pm, and a closing session from 3.30 pm to 4 pm. They discover that returns “tend to be easier to predict in the middle of the day rather than at the beginning of the day, possibly due to large moves typically [being] concentrated during the overnight hours,” while durations are similarly predictable.
Despite the higher volatility, they note that “interestingly,” “predictability at the end of the day is higher for all of the prediction issues we investigated,” “suggesting that the trading patterns at the end of the day are more constant across days.”
But predictability quickly approaches zero. The team discovers that after just five minutes, or roughly 2,000 transactions, “prediction becomes no better than a coin flip.”
The effects of a delay and a look forward
Other elements also influence the precision of HFT predictions. According to researchers, the most recent 10 milliseconds, or 10 transactions, account for 80% of the overall predictability. However, they also demonstrate that adding a little artificial latency to data processing ‘decreases sharply’ predictability.
The IEX Stock Exchange uses this “speed bump” concept to generate a 350-microsecond delay by adding an extra cable to its message system. The exchange claims that the idea behind the speed bump is that no market player should be able to move more quickly than the exchange itself.
The research also examines how “signal” affects order flows from various exchanges. We replicate the effect of having an (imperfect) glimpse at the incoming order flow, a look ahead ability that is frequently credited to the fastest HFTs, in terms of boosting the predictability of the subsequent returns and durations, according to the team’s report. They discover that this ‘peek’ increases price direction accuracy from 68% to 79% and return predictability from 14% to 27%. The researchers emphasize that they “take no position on whether traders have in actuality the ability to predict the direction of the incoming order flow.” Still, they assert that “what is evident is that such ability is (or would be) tremendously beneficial.”
The Princeton team describes the goals of the publication and what the research is not intended to accomplish. In their statement, they state that “this research is about quantifying the predictability in high-frequency short horizon prices, illustrating how it may be done, and evaluating the impact of diverse contexts.” The question of why there is such regularity in the first place.
The “how and when” examination in this study is a logical pre-requisite but answering the “why” question is undoubtedly intriguing as well. Furthermore, a whole different set of techniques than those used in this study would be needed to understand the underlying causes of predictability, which range from particular features of the architecture and operation of financial markets to traders’ strategies.
From January 2019 to December 2020, which encompassed the onset of the Covid-19 pandemic and what researchers refer to as the “extremely volatile environment of March and April 2020,” the team used “the complete transactions and quote data for the 101 equities that were constituents of the S&P 100 index.”