Does Fragmentation Destroy Depth?

A trader pulls up the order book for a large-cap US stock on a single exchange and sees 1,200 shares at the best bid. A colleague, checking a consolidated feed that aggregates every venue the stock trades on, sees 9,400 shares at that same best price. Both are looking at the same stock at the same instant; the difference is that US equities trade across more than a dozen exchanges and dozens of dark pools simultaneously, and any one venue only shows its own slice. The question this raises — does splitting order flow across so many venues actually reduce the liquidity available to trade, or does it just scatter the same liquidity across more places? — has a more nuanced answer than either "fragmentation is obviously bad" or "it makes no difference."

Depth on one book versus depth across the market

Define local depth as the size resting at the best price on a single venue, and consolidated depth as the sum of size at that price across every venue trading the stock. Before Regulation NMS-style fragmentation, local depth and consolidated depth were close to the same number, because almost all order flow concentrated on one exchange. After fragmentation, local depth on any individual venue shrinks — but if the total pool of resting orders hasn't shrunk, only redistributed, consolidated depth should be roughly unchanged. Empirically this is close to true for large, liquid names: research following Reg NMS found that consolidated NBBO depth held up reasonably well even as any single exchange's own book thinned considerably, because smart order routers stitch the fragments back together for anyone actually trying to trade size.

The catch is that consolidated depth is only useful to a trader who can actually access all of it fast enough and cheaply enough. A trader who only routes to one venue experiences the fragmented, thinner world; a trader with a smart order router that can sweep all venues simultaneously experiences something closer to the consolidated picture, minus the latency and fee costs of routing to many places at once.

A worked example

A stock has a genuine 6,000 shares of buyer interest at the best bid. On a single-exchange market structure, all 6,000 shares would sit on one book, and a seller could hit the whole thing with one marketable order (transaction costs aside). Under fragmentation, suppose that interest splits as 1,500 shares on Exchange A, 1,100 on Exchange B, 900 on Exchange C, 700 in a dark pool, and the remaining 1,800 scattered across nine smaller venues in lots of roughly 200 shares each. A seller who only routes to Exchange A sees 1,500 shares of liquidity — apparently a quarter of what existed before fragmentation. A seller using a smart order router that can hit all fourteen venues within a few milliseconds captures close to the full 6,000 shares, paying only a small amount of extra cost in exchange fees and the latency of routing to more destinations. The liquidity didn't vanish; it became inaccessible to anyone without the technology to reassemble it.

Where this gets used

• Smart order routing design: routers are built around the empirical fact that meaningful size exists off the primary listing venue, and their whole value proposition is capturing consolidated rather than local depth.
• Regulatory debate: whether fragmentation harms "market quality" hinges on this distinction — critics point to thinner individual books and more venues to monitor, defenders point to stable or improved consolidated spreads and depth for the names studied.
• Small and illiquid names: the reassembly argument is weakest here — with genuinely little total liquidity to begin with, spreading it across many venues can leave every individual book too thin to trade against without real impact, unlike the large-cap case above.

In interviews

If asked whether fragmentation is good or bad for markets, avoid a one-line verdict. Distinguish local depth from consolidated depth, note that for liquid names the total usually survives fragmentation intact while access to it becomes technology-dependent, and flag that this breaks down for illiquid names where there wasn't much depth to redistribute in the first place. That three-part answer is what separates a mechanistic understanding from a talking point, and it sets up Smart Order Routing as the practical fix for the access problem.