Math From Biology Finds Warning Signs In Tokyo Stock Trading

Stock Markets May Show Symptoms Days Before Turmoil, Study Finds

Stock market swings can seem to arrive out of nowhere. A new study found hints that trading data may have been waving a flag first. Days before several of the Tokyo Stock Exchange’s wildest swings in 2020, a telltale shakiness showed up in how groups of anonymous market participants were trading.

Researchers borrowed a method originally used to catch disease before symptoms appear, called Dynamical Network Markers, and applied it to real stock market trading data. In biology, the method looks for groups of genes that grow erratic just before a bigger breakdown. Here, the groups were traders.

Published in EPJ Data Science, the study is an early look at an idea, not a tool for timing the market, and the pattern it found went muted during the sudden COVID-19 market shock.

Stock Market Early Warning Signs Emerged From Sorting Anonymous Trading Into 186 Groups

Records from the Tokyo Stock Exchange hide who sits behind each order, but they do include coded ID numbers for the computer systems, called virtual servers, that submit them. Those records run from November 2019 through 2020, a stretch that includes the COVID-19 market chaos of February and March.

Rather than treat each server as its own trader, the team grouped servers that behaved alike into 186 participants: 68 high-frequency traders, which place and cancel orders at lightning speed, 25 brokers, 3 general investors, and 90 in a catch-all others category.

For each participant, the team tracked trading volume, how much its trading overlapped with others in the same stocks at the same moments, and how bunched into bursts its order timing was. Shakiness meant how much those measures bounced around during a trading session, averaged over about a week, much like a pot lid rattling before the water boils over.

Market swings were measured with the Tokyo Stock Price Index, a standard gauge of the overall market. Days when prices swung unusually hard compared with the day before turned up eight times. Four fell during the COVID-19 shock, so the team kept only the first, leaving five turmoil days: January 8, January 27, February 25, June 16, and August 28.

To test the idea, researchers used four turmoil days to pick the participants whose trading grew shakiest beforehand, then checked whether those same participants grew shaky before the fifth day. It works a bit like training a smoke detector on four past fires and testing it on a fifth.

Early Warning Signs Rose Before Several Turmoil Days but Stayed Muted During the COVID-19 Shock

Across various kinds of trading data, shakiness tended to climb in the days before several turmoil days, especially January 8. According to the researchers, that rise likely reflected traders adjusting ahead of a scheduled exchange date on January 10, which the method flagged without being told the calendar. By contrast, a standard measure based on recent price swings rose only after turmoil arrived, in the cases checked.

During the COVID-19 shock, price swings spiked repeatedly while the shakiness measure stayed comparatively smooth. Possible explanations include fast-moving outside news driving those swings, a measure that averages a week of trading and may react too slowly, and some spikes arriving after the market was already unstable. The paper presents these as possibilities, not conclusions.

No single participant’s signal held up under statistical testing across the full study period. Information about instability may live in the collective behavior of groups rather than in any one trader.

Two of the Repeat Traders Sat at the Center of the Trading Web

Several participants kept turning up among the shakiest across different turmoil days and types of data, and the researchers labeled them principal participants. They did not stand out for placing the most orders or trading the largest volumes. Two did stand out another way: a broker and a participant from the others category sat in unusually central spots in a web linking participants that traded the same stocks at the same time. Other repeat participants sat in less central positions, so network position alone does not explain who kept showing up. Authors say well-connected traders may be associated with how stable the market stays, a possibility future work would need to test.

Early Warning Signs in Past Data Are Not a Forecasting Tool Yet

Spotting a pattern in old records is different from predicting new events. In the paper’s words, the results “do not establish a highly accurate forecasting framework.” Building an alert system would require working out how much lead time the signal gives and what level should trigger a warning.

Still, a method built to catch disease early turned up warning patterns in a very different setting. If further testing backs them up, the first hints of market trouble may show up in how traders behave, days before prices lurch. Someday that could give regulators and risk managers a head start, though that day has not arrived.

---

Disclaimer: This article summarizes early-stage research for general information only. It is not financial, investment, or professional advice. Readers should consult a qualified financial professional before making investment decisions.

---