On Oct. 5, Nasdaq told its market-data customers to prepare for a 23-hour trading day. Three days later, the Securities and Exchange Commission published a Nasdaq rule filing addressing how trading would resume after a market-wide circuit-breaker halt under that expanded schedule. The industry is targeting Dec. 6 for broader overnight trading. Wall Street is rebuilding the machinery for a market that barely sleeps.
There is another machine that needs attention: the automated trading strategy. If its price information turns stale at 2 a.m., who or what takes away its permission to increase somebody else’s risk?
This is no hypothetical market waiting to open. Alternative trading systems already execute U.S. stock orders overnight. An SEC staff analysis found that August overnight trades accounted for just 0.9% of total share volume but still averaged 144.6 million shares and $7.4 billion per trading day. Ten stocks accounted for 43.4% of overnight volume, versus 10.3% during regular hours. That is a thinner, more concentrated market, not a miniature version of midday Wall Street.
The data problem is real, too. The SEC found that 21.2% of overnight share volume reported by alternative trading systems occurred from 8 to 9 p.m. Eastern, the hour designated for maintenance in the planned consolidated-data schedule. Other venue or proprietary feeds can remain available. But an open market is not a guarantee that every reference feed, quote, and price check is current.
Consider a brokerage algorithm reacting to a surprise earnings report. It wants to double a position after an apparent price move. One feed is delayed, spreads are widening, and another market shows a different quote. The program may still be operating exactly as written. The evidence behind its next order is weaker than the evidence behind the last one.
The rule I would put in the brokerage’s controls is simple: When a critical data condition fails, an automated system should lose the permission to increase net exposure. It might still be allowed to reduce a position, preserve an existing properly priced limit order, or perform other preapproved defensive tasks. The exact permissions depend on the strategy and the market. But taking more risk should require fresh qualifying data or an authorized override.
That is not a demand to shut down trading every time a feed flickers. A blanket halt can damage liquidity and prevent customers from reducing positions. This is a narrower protection: Make authority contract as evidence deteriorates, instead of leaving the same risk-taking permission in place until an operator notices.
Existing market protections are necessary, but they address different questions. The SEC has backed overnight price bands, and exchanges are updating circuit-breaker rules. FINRA already requires extended-hours supervision, best-execution reviews, and customer risk disclosures. Those rules help protect market orderliness and investors. A market-wide price band does not establish whether one firm’s algorithm should enlarge a customer position after losing a trusted reference price.
Nor should anyone pretend that fully autonomous generative-AI traders already dominate exchanges. The Financial Stability Board reported little evidence of generative AI being used for fully autonomous trading. Conventional algorithms can already route orders and alter exposures without a person approving each step. The risk-control requirement should apply to those systems now, whatever their marketing label.
Brokerages can act without waiting for a new Washington regulator. Before enabling an automated strategy overnight, define its permitted securities, position and order limits, acceptable quote age, independent price references, and the conditions that block new exposure. Put the enforcement outside the trading model so that the software cannot waive its own limits. Test the system with stale quotes, missing feeds, conflicting prices, and orders already queued when a limit changes.
The test should record how long the firm took to withdraw risk-increasing permission, whether queued orders were canceled or constrained, and whether fresh data actually justified any restoration. Supervisors should be able to reconstruct the active data sources, model and policy version, limits checked, and any human override. A green dashboard in the morning is not evidence that controls worked overnight.
Investors deserve something more useful than a generic warning about thin liquidity. Brokerages could publish understandable session-specific execution-quality information, including spreads, fills vs. available contemporaneous quotes, and unfilled orders, with clear caveats where price comparisons are incomplete.
SPAIN IS BECOMING EUROPE’S CUBA — AND AMERICAN CAPITAL IS THE NEXT TARGET
The transition to longer trading hours is worth making. It can serve overseas investors and people who cannot trade during the conventional workday. But market access and permission to assume risk should not be confused. A broker can offer the first while tightly governing the second.
At 2 a.m., the market may still be open. An algorithm’s authority to increase someone else’s exposure should remain conditional.
Burak Oktenli is a graduate student in applied intelligence at Georgetown University and an independent researcher focused on trustworthy artificial intelligence, cybersecurity, autonomous systems, and high-consequence technology governance. He holds a bachelor’s degree in computer science and engineering from the University of South Florida and a master’s in business administration.
Continue reading...
[ H/T Washington Examiner ]
