The Swiss spread advantage disappears on comparable electronic venues. Across 46,016 observations, the coefficient for Swiss CLOB trades against U.S. ATS trades is -0.0119, with a standard error of 0.0096. Mattmann highlights the same result in the introduction: pricing converges when the trading protocols resemble each other. The entire advantage appears in bilateral trading, where the CHF dummy is -0.3051 percentage points (s.e. 0.0115, n=362,423). Both coefficients come from a single regression of effective bid-ask spreads on dealer-to-client trades, separated by channel. Spreads are the only measure divided this way. The paper is most vulnerable at precisely this split.
Listed bonds, OTC trading
Switzerland gives Mattmann an unusual setting. Nearly all CHF-denominated corporate bonds are listed on SIX, where a continuous central limit order book displays executable quotes. Institutional-size trades can instead go through a parallel OTC market, with reports reaching the exchange within fifteen minutes. During the sample year, dealers also had to route client orders below CHF 100,000 to the book. The obligation ended on 1 February 2019. Displayed pre-trade prices and bilateral negotiation therefore coexist for the same bonds at the same time.
Mattmann matches proprietary SIX transactions and quotes covering 1,070 bonds and 267,009 trades, along with daily average bid-ask quotes and quoted volume, to enhanced TRACE data covering 17,024 bonds and 14,869,575 trades. Both samples cover calendar 2018. Price dispersion follows Abudy and Wohl. The Amihud ratio follows Dick-Nielsen and coauthors, excluding retail, while effective spreads follow Choi and coauthors and use the daily volume-weighted interdealer price as the benchmark. SIX lacks trade direction, so the spread measure is the absolute deviation from that benchmark. The regressions compare a CHF dummy after interacting issue-size, age, maturity, rating and month fixed effects, with clustering by bond and month.
Median dealer-to-client effective spreads are 0.16% in Switzerland and 0.38% in the U.S. Price dispersion is 0.06% against 0.25%. The median Amihud ratio is 0.10% against 0.27%, again using dealer-to-client trades aggregated to the bond level. After fixed effects, CHF bonds show spreads lower by 0.3701 pp (s.e. 0.0096), dispersion lower by 0.2798 pp and Amihud lower by 0.4978 pp.
Trading activity runs the other way. Median monthly turnover is 1.3% against 2.2%, with 12 trades per bond-month against 28 and trading on 35% of days against 57%. A market recording 12 trades a month per bond rather than 28 still prices at 0.16% rather than 0.38%.
The activity-equals-liquidity reflex fails here.
Can displayed quotes anchor a 500,000 OTC negotiation?
The spillover case requires shallow displayed quotes to influence transactions far beyond displayed depth. Median quoted volume for each CHF bond is about 113,000. The median Swiss OTC trade is 500,000. Under the paper's Hendershott-Madhavan categories, that remains an odd-lot, far short of block size at 5m and above.
Mattmann's direct evidence comes from execution location. Of Swiss dealer-to-client OTC trades, 95% print within the average daily CLOB bid-ask range, while 63% fall closer to the quoted midpoint than to either quote. A bond-day regression then relates OTC effective spreads to quote quality. With the full fixed effects, the quoted CLOB spread has a coefficient of 0.3565 (s.e. 0.0687), n=5,213, R-squared 0.56. Log quoted depth enters at -0.3078 by itself, then drops to -0.1221 (s.e. 0.0569) when both variables appear.
Quoted spreads and negotiated prices plainly move together. Their sequence remains unknown. Each could reflect the same unobserved bond liquidity, leaving the direction unidentified. Mattmann states the limitation directly: these are cross-sectional associations and they do not establish a causal relationship. The actionable question rests on the more cautious interpretation.
The Swiss size gradient is too noisy
The paper gives one result more weight than its standard errors warrant. U.S. corporate bonds have a well documented size-cost gradient, which appears in this sample. Among OTC transactions, spreads for odd-lot, round-lot and block trades are 29.70, 46.10 and 45.36 bps below micro-sized retail trades, all at p<0.001.
Swiss OTC size categories are insignificant, which the paper interprets as weaker price discrimination. Yet the Swiss odd-lot estimate is -0.2983, with a standard error of 0.2259 across 9,115 observations. Its magnitude is roughly comparable to the U.S. effect. Precision is another matter: the standard error is 0.2259 versus 0.0014 in the U.S. column, about 160 times wider. The interval readily includes the U.S. gradient. It cannot separate a flat cost curve from an ordinary gradient estimated on 9,115 Swiss OTC observations rather than 6,496,206 U.S. observations.
The benchmark creates another problem. Effective spreads use the daily interdealer VWAP. The reported channel shares show 85% of all Swiss transactions and 21% of volume executing in the CLOB, with dealer-to-client and interdealer trades combined. We read those shares as evidence that the Swiss benchmark price and treatment are not independent. For a market whose median bond sees 12 monthly trades and trades on 35% of days, the reference price draws on very few prints. Any resulting bias has no obvious direction.
One concern can be sized and set aside. SIX fees fall outside the sample: CHF 1.5 per transaction, plus a value fee of up to 0.01% on client orders and capped at CHF 100. On the median 25,000 order-book trade, the charge is under two basis points. It comes nowhere close to the 22 bp median spread gap. The paper does not address bank commissions, which would not appear in transaction prices reported to SIX.
The study design imposes the larger constraint. It compares one calendar year across two countries whose markets differ by fifty times in outstanding value, CHF 207bn against USD 10,590bn. Median ratings are A and BBB+, while high yield represents 3% against 20%. Within the Swiss sample, 20% of bonds have no rating. Some of the rated issues rely on Zurich Cantonal Bank rather than S&P or Moody's, and the fixed effects reduce ratings to four coarse buckets. Domestic pension demand for CHF paper and buy-and-hold behaviour receive no controls. Either could lower price dispersion without a transparency channel.
Mattmann identifies the cleaner test as future work. Removal of the CHF 100,000 routing obligation on 1 February 2019 creates a within-market pre/post comparison. A subsequent widening of Swiss OTC spreads would move me a long way.
Publish the quotes, leave large trades bilateral
An executable reference price that remains continuously visible appears to influence negotiations conducted elsewhere. Chen and Zhong estimate the U.S. counterfactual at $846 million a year in transaction-cost savings from NYSE pre-trade transparency. Mattmann makes a narrower claim for his contribution, and it holds up: he is the first to formally compare an OTC-driven market with one where a centralized exchange operates alongside OTC trading.
The defensible policy claim has the same scope. Publish live executable quotes broadly, including for participants who never intend to trade on the venue. The paper's conclusion rejects the stronger inference that every corporate bond transaction should migrate to an order book, stating that the results do not imply fully centralized exchange trading is optimal for all corporate bond transactions. Although 85% of Swiss transactions use the book, they represent only 21% of volume. Large trades remain bilateral in both markets.
For anyone trading CHF paper, the execution numbers matter. The median quoted CLOB spread is 0.56%, while the median implied effective spread against the quoted midpoint is 0.26%. Trades execute at roughly half the posted spread, meaning the displayed touch systematically overstates the price paid.
We could not test any of this ourselves. Testing the mechanism requires executable bond quotes with quoted depth and OTC transaction prices carrying direction. We hold no corporate bond transaction or quote data, neither TRACE nor SIX. Substituting bond ETFs would remove the object being measured: the interaction between a displayed book and a simultaneous bilateral negotiation in the same instrument.