Japan Exchange Group published the English version of a working paper on Friday finding that dark pool trading tends to fall during volatile periods and can hurt market quality when volatility is high.
The paper, originally released in Japanese on January 30, was authored by Tokyo Stock Exchange equities manager Hiroaki Wakamatsu. It uses the TSE’s dark pool flagging system, introduced in June 2020, which identifies orders matched in dark pools and subsequently routed to the ToSTNeT market.
The study finds that dark pool trading is conditionally helpful. During low-volatility environments, it has a positive effect on market quality, providing additional liquidity and price improvement without disrupting price discovery. During high-volatility periods, the effect reverses.
The paper finds that institutional investors, who remain the dominant users of dark pools, frequently reduce their use of these venues during stressed markets to avoid execution lag. This drives down the “dark pool ratio,” the proportion of dark pool trading value relative to the TSE auction market, during volatile periods.
Price improvement depends on order book conditions
Wakamatsu’s analysis also examines execution prices in dark pools compared to the TSE auction market. Price improvement rates are heavily influenced by the state of the order book, specifically the “micro-price,” a benchmark that incorporates order book depth rather than just the midpoint.
Self-execution trades, where the broker is on the other side of the order, generally show weaker price improvement rates for the client, the paper finds.
The research comes as dark pool trading in Japan has expanded sharply. TSE data shows the equities dark pool ratio rose from 2.78% in September 2020 to 5.44% in September 2025, with average daily trading value climbing from ¥112 billion to ¥487 billion over the same period.
Retail participation is also growing as online securities firms increasingly use Smart Order Routing to send individual investor orders to dark pools. Institutional flow, however, still drives the volatility-sensitive behaviour that the paper documents.
