2.5.2 Fragmentation and the NBBO

Basic Prerequisites: 01 - From Zero and 03 - Exchanges & Venues.

---

1. Intuition & Practical Objective

Fragmentation creates an information problem: if ten venues each publish their own quote, who publishes “the” price? The answer is the NBBO - the National Best Bid and Offer - the best bid and the best ask taken across all venues, plus the consolidated tape, the merged stream of every venue’s trades. SOR cannot exist without them: they are the router’s eyes.

This page makes three things precise:

1. The NBBO as a cross-sectional extreme. The best ask is ${\min }_i{a}_i$, the best bid is ${\max }_i{b}_i$ - not the quote you happen to be looking at. A trader whose broker watches only its own venue is structurally looking at the wrong price.
2. The consolidated view is imperfect. Venues report at slightly different speeds, so the NBBO is a lagged composite: quotes that are locked or crossed (a bid ≥ another’s ask) exist momentarily, and reading a stale NBBO is the seed of the trade-through failure (page 05).
3. Fragmentation redistributes price discovery. When one stock trades on many venues, where the price is discovered becomes an empirical question, answered with cointegration and information shares (Hasbrouck Ch 10).

The one-sentence essence. “The NBBO is the market’s best bid and offer formed by taking the extreme quote across every venue, while the consolidated tape merges every venue’s trades - together they turn a fragmented market back into a single, viewable price, at the cost of some latency.”

---

2. Mathematical Ground Truth & Derivations

The NBBO. With venue best asks $a_i$ and bids $b_i$,

$$\boxed{a^{\ast }=\underset{i}{\min }{a}_i,b^{\ast }=\underset{i}{\max }{b}_i,\text{mid }m=\frac{1}{2}(a^{\ast }+b^{\ast }),\text{spread }{S}^{\ast }=a^{\ast }-b^{\ast }.}$$

A locked market is $a^{\ast }=b^{\ast }$; a crossed market is $a^{\ast }<b^{\ast }$ (some venue’s bid exceeds another’s ask) - both are transient artifacts of reporting latency across venues.

The consolidated-tape fragmentation index. Let venue $i$ execute volume $v_i$ over a window, with shares $s_i=v_i/{\sum }_j{v}_j$. The Herfindahl concentration index

$$\text{HHI}=\sum _i{s}_i^2,\text{effective number of venues}=\frac{1}{\text{HHI}},$$

so HHI $\to 1$ is a single monopoly venue and HHI $\to 0$ is a perfectly fragmented market. O’Hara & Ye (2011) proxy fragmentation in individual stocks using newly-available TRF (trade reporting facility) volumes - trades in off-exchange ATS/dark venues - precisely because that volume is invisible in the lit quotes.

Effective vs quoted cost - the cost of a fragmented view. The cost a taker actually pays is measured against the contemporaneous mid (Foucault Ch 2):

$$S_e\equiv d(p-m),d=\{\begin{array}{ll}+1& \text{buy}\\ -1& \text{sell}\end{array},$$

the effective spread. If a router buys at venue $i$‘s ask $a_i$ while a better ask $a_j<a_i$ rests elsewhere, the excess $a_i-a_j>0$ is pure routing loss - the gap the NBBO exists to eliminate.

Price discovery across venues (Hasbrouck Ch 10). Prices of the same security on different venues are cointegrated: they share one common efficient price $m_t$,

$$p_t=m_t\iota +s_t,\iota =(1,\dots ,1{)}^{\prime },$$

where $s_t$ is stationary (the cross-venue spread/basis). A cointegrated vector moving average is non-invertible, so one cannot fit a convergent VAR in first differences; instead fit a VECM

$$\Delta p_t={\varphi }_1\Delta p_{t-1}+\cdots +\beta (z_{t-1}-b)+{\epsilon }_t,$$

and decompose ${\sigma }_w^2=[\theta (1){]}_1[\theta (1){]}_1^{\prime }$ (the common random-walk variance). Each venue’s information share is its relative contribution $d_i^2/{\sigma }_w^2$ to the common efficient-price innovation, reported as a min–max band over the (Cholesky) orderings. This is how you answer “does the price get discovered on the primary exchange, on the fast ECN, or in the dark?”

Empirics (O’Hara & Ye 2011). Across matched fragmented/consolidated stocks post-Reg NMS: effective spreads are lower in the fragmented sample by 0.29 cents (median 0.11 cents), execution speed differs on the order of 7 seconds, and realized spreads are statistically indistinguishable - i.e. fragmentation lowers the trading cost to the taker without changing liquidity suppliers’ profit. Fragmentation raises short-term volatility but makes prices more efficient (closer to a random walk).

---

3. Computational Implementation - build an NBBO from fragmented quotes

Consolidate venue top-of-books into the NBBO, measure the cost of a stale single-venue view, and compute a fragmentation index. Standard library only; re-executed and reproduced.

Read the numbers. The NBBO bid (100.00) rests on venue B and the NBBO ask (100.01) on venue C - two different venues, neither of which is the “obvious” one. A trader watching only venue B buys at 100.03, 2 cents worse than the true best ask; a trader watching only A pays 1 cent worse. The Herfindahl index of 0.2850 says the market behaves like 3.51 equally-sized venues - highly fragmented. The whole job of the consolidated view is to erase those 1–2 cent gaps before the order is sent.

---

4. Failure Modes & First-Principles Breakdowns

1. Single-venue blindness. Using one venue’s top-of-book as “the price” costs 1–2 cents per share in this example - 1–2 bps per side, 2–4 bps round-trip on a stock trading near 100.00 with a 0.5 bps NBBO half-spread.
2. The NBBO is a lagged composite, not a fact. Venues report at different speeds; quotes lock and cross transiently. A router that trusts the NBBO without re-checking at execution can trade through a better protected quote (page 05). The NBBO is best treated as a prediction of the venue’s state, not a guarantee.
3. Fragmentation ≠ worse quality, but ≠ free either. O’Hara & Ye (2011) find spreads fall and speed improves with fragmentation, but the infrastructure cost (consolidated feeds, latency) is real, and each thinner book means a bigger fraction of your order walks to the next venue. See 05 · Failure Modes.

---

5. References

• Hasbrouck, Joel - Empirical Market Microstructure (2007)
• Hasbrouck, Joel - “One security, many markets: Determining the contributions to price discovery,” JF 50(4), 1995. The origin of the information-share measure used above.
• O’Hara, Maureen & Ye, Mao - “Is market fragmentation harming market quality?” JFE 100(3), 2011.
• Foucault, Pagano & Röell - Market Liquidity (2013)

---

6. Connected Graph Bridges

• Back: 01 · From Zero · Index Hub
• Continue: 03 · SOR Logic
• Venue map: Exchanges & Venues
• Spread decomposition: Spread Decomposition & the Roll Model