Competition and Exclusivity in OTC Markets

Rainer Haselmann, Ingomar Krohn, Philippe Mueller, and Maik Schmeling

Discussion by Chaojun Wang

The Wharton School
University of Pennsylvania

CDI Conference on Derivatives

September 2026

MDP has the potential to greatly intensify competition

Dealer 1 Dealer 2 Dealer 3 customer p1 p2 p3 Traditional OTC Trading Dealer 1 Dealer 2 Dealer 3 MDP customer p̃1 p̃2 p̃3 p̃1, p̃2, p̃3 Multi-Dealer Platforms

Costs double from one dealer to two, then fall

TCi = α + ∑k ∈ {2, 3–4, 5–10, 11+} δk 1{Di = k} + γ′Xi + FE + εi

Mean transaction cost (bps) 0.65 1.31 0.58 0.36 0.28 ×2.0 1 2 3–4 5–10 11+ Number of dealers invited (D) D = 2 premium δ2 (bps) raw + pair FE + provider FE + requester FE + req × prov FE 0.65*** 0.59*** 0.41*** 0.00 0.07** absorbed by client FE
Mean costs: Table III, all pairs. δ2: Table IV cols. (1), (2), (4), (3), (7); rows 2–5 add pair FE, log notional and bid–ask spread.
t = 5.33, 5.19, 5.15, 0.10, 2.09, two-way clustered by requester and provider. 1.33m trades, 27 pairs, Jan 2018 – Jul 2022.

A second dealer helps narrow, hurts broad clients

TCi = ∑k ≠ 1 δkg 1{Di = k} + γ′Xi + αc,m + αr,p + εi (one regression per group g)

Mean transaction cost (bps) D = 1 D = 2 2.08 1.25 0.46 0.53 1.31 1.83 1.21 0.83 Narrow ≤5 Mid 6–9 Broad ≥10 Instit. Requester group δ2 within client–dealer pairs same client, same dealer, over time Narrow ≤5 Mid 6–9 Broad ≥10 Instit. 0 −0.159*** 0.068 0.141*** 0.043
Equation (5): αc,m = pair × month FE, αr,p = requester × provider FE, estimated separately for each requester group.
Means: Table V (change −0.77***, +0.58**, +0.75***, +0.29**). δ2: Table VII (s.e. 0.056, 0.076, 0.048, 0.061). 1.32m trades.

Broad clients are quoted tighter only when D = 1

QCi,p = αr,p + αc,m + αh + γ RFMi·1{Di = 1} + ∑k βk 1{Di ∈ bink} + δ′Xi + εi,p

Quoted cost vs. bilateral (bps) dealers never see D 0.58 0.49 0.37 0.31 2 3–4 5–10 11+ Number of dealers invited (D) β2 within client–dealer pairs same dealer alone, then one rival Narrow ≤5 Mid 6–9 Broad ≥10 Instit. 0 −0.050 0.348*** 0.924*** 0.650***
Equation (9) on 12.1m direction-aligned quotes, with requester × provider, pair × month and time-of-day FE. D is never disclosed.
Table XI col. (1); Table XIII cols. (1), (4), (7), (10), s.e. 0.088, 0.109, 0.203, 0.129. Bilateral is the reference bin throughout.

1. More detailed institutional settings

The paper rests on two features of the protocol

Every solicited dealer quotes

D is never disclosed to any dealer

Three things I would like to know

  1. Can a client bargain bilaterally over the price with a dealer of its choosing?
  2. How long does a typical auction last?
  3. What determines which dealers are available to a client?

Requester interface on MarketAxess for U.S. corporate bonds

Request-for-quote inquiry screen with a dealer list on the right allowing many dealers to be selected at once

2. Trade protocol matters

SEF discloses D to invited dealers.

Customers mostly contact only 3 dealers, the minimum required by regulations.

Dealer response rate is around 90%.

(Riggs, Onur, Reiffen, and Zhu, 2020)

MarketAxess does not disclose D to any dealers.

Customers contact 25+ dealers on average in an RFQ.

Dealer response rate is around 25%.

(Hendershott and Madhavan, 2015)

2. More dealers ↓competition when dealers can ignore a request

Model prediction (Wang, 2023 JFE) Transaction cost lowest 1 2 3 4 5 6 7 8 9 10 D The paper's data Transaction cost (bps) 0.65 1.31 0.58 0.36 0.28 1 2 3–4 5–10 11+ D

External validity: Does the exclusivity discount hold if dealers can ignore RFQs?

If so, the SEF minimum of 3 dealers is a policy mistake that worsens competition.

How does a dealer adjust quotes without observing D?

QCi,p = β X̄r,pW,t + αi + εi,p (αi = request FE)

Dealer’s record with the client prior 90-day mean, 95% CI price impact (per bp) realized spread (per bp) competitive share (0–1) −0.002 0.009** 0.757*** 0 0.5 1.0 Coefficient (bps) How large is the inference? implied quote change (bps) competitive share 0 → 1 D = 2 vs D = 1 D ≥ 11 vs D = 1 0.76 0.58 0.31
Left: equation (10) at the 90-day window, Table XII cols. (1), (4), (7). Price impact and realized spread are per bp; competitive share is a 0–1 share.
Right: Table XII col. (7) against Table XI col. (1). Longer windows give 0.791** and 0.736** for competitive share.

3. More corroborative tests for the inference mechanism

  1. Can past D predict current D?
  2. Does a client tend to contact the same set of dealers over time?
  3. After a client rejects a bad quote from a dealer, does the client immediately make a request from a broader set of dealers for the same currency pair?
  4. If a client receives a bad price from a dealer, does the client punish the dealer by excluding it from future requests, especially when D = 1?

4. A simple model rationalizing the paper’s finding

Client seeks to sell one unit of the asset. The client values the asset at 0.

Dealer i values the asset at vi ~ F.

D ≥ 2: First-price auction with private values {vi}1 ≤ i ≤ D.

b(v) = v − F(v)1−D ∫0v F(x)D−1 dx,  E[maxi b(vi)] = E[v(2)]

D ≥ 2: b(v) and E[maxi b(vi)] both ↑ with D

D = 1 (take-it-or-leave-it offer): The dealer’s quote should make the client indifferent between using the same dealer and inviting all dealers that it has access to in future requests. This indifference condition determines the dealer’s equilibrium bid.

D = 1 (bargaining): The gains from trade between the dealer and the client are higher when D = 1 because the dealer is guaranteed to execute the trade. The dealer and the client split that surplus, which determines the equilibrium trading cost for the client.

Summary

Exclusivity discount on multi-dealer platforms is a clear contribution.

I suggest

  1. Discuss the external validity of the finding on platforms whose institutional settings differ slightly from this one. If it holds, the paper delivers a concrete policy prescription.
  2. More corroborative tests on the inference mechanism.
  3. Jointly model the exclusivity discount for D = 1 and competition for D ≥ 2.