OTC Discount

1. Introduction

Most fixed-income and derivative instruments trade in over-the-counter (OTC) markets. Participants in these markets search for counterparties and negotiate prices that reflect their relative bargaining positions. Instead, in some asset classes, dealers trade with one another on electronic central limit order books provided by exchange platforms. In contrast to OTC markets, search-and-bargaining frictions are irrelevant on the exchange where all participants can immediately trade at the same prices. In practice, OTC and exchange platforms often coexist as hybrid markets.

The dealer-to-dealer (D2D) segment of the market for German sovereign bonds (Bunds) is a case in point. We document that the vast majority of OTC Bund trades have lower transaction costs than those implied by exchange quotes, thus giving rise to a pervasive OTC discount. A substantial share of these OTC transactions is intermediated by interdealer brokers. Distinguishing between bilaterally negotiated and broker-intermediated OTC trades, we show that the latter features a considerably lower OTC discount. We thus observe a pecking order of transaction costs; exchange trading is more expensive than via brokers, and broker-intermediated trades are more costly than bilateral OTC trades.

These findings are difficult to square with theories that exclusively focus on search-and-bargaining frictions. If search was the only friction, then traders with access to an exchange should not have a reason to trade bilaterally. That said, the existence of a hybrid market and our finding of an OTC discount are consistent with theories emphasizing the role of asymmetric information, which gives rise to adverse selection. In line with this view, we document that dealers’ choice of trading protocol is driven by variables measuring the private and market-wide information that they have access to. Moreover, we show that the pecking order of transaction costs is aligned with the informational content of order flows across protocols for Bund returns. Finally, our results imply that the observed variation of transaction costs within OTC protocols is driven by information as well as search-and-bargaining frictions. Combined, our analysis suggests that a realistic description of hybrid markets requires both types of frictions. This is relevant given recent efforts to regulate OTC markets more tightly and to foster exchange-based trading.

In the interdealer Bund market, dealers trade with each other using three different protocols: via the interdealer exchange MTS (Mercato Telematico dei titoli di Stato), via interdealer brokers, and through bilateral negotiations. Trading on the exchange is immediate and post-trade transparent. As such, it eliminates search-and-bargaining frictions but can also reveal private information because all market participants instantaneously observe the price and volume of each trade. In contrast, for bilateral and broker-intermediated OTC transactions, only the involved counterparties observe a trade. In bilateral OTC transactions, dealers search for counterparties and therefore, know their identities. Instead, when they trade via a broker, this search effort is undertaken by the broker, and counterparties remain anonymous to one another. For each trade, dealers can thus choose between three protocols, which differ with respect to anonymity, transparency, and immediacy. Hence, their trading decisions depend on the relative importance of search-and-bargaining frictions as well as information frictions across protocols. What the presence of these frictions implies for the ranking of transaction costs is an empirical question.

The D2D Bund market represents an ideal laboratory to study this question. We combine a unique regulatory data set that comprises all trades involving at least one financial institution regulated in Germany with the full limit order book on MTS. This allows us to compute our main quantity of interest, OTC discount, which measures the transaction cost advantage of an OTC trade with respect to the potential execution of the same trade on the exchange at the same time.¹ We find that 87% of interdealer transactions execute at a lower transaction cost than the one attainable on MTS. Moreover, there is a sizable difference between the OTC discounts of bilateral and broker-intermediated OTC trades. Although bilateral OTC trades on average feature 54.6% lower transaction costs relative to exchange quotes, broker trades are on average only 24.6% cheaper. This pecking order of transaction costs is highly economically significant in light of an annual trading volume of about 1.6 trillion Euro in the interdealer Bund market.

More than three quarters of interdealer trading volume in the Bund market is intermediated by brokers. Why do dealers rely so heavily on brokers despite the higher transaction costs relative to bilateral OTC trades? Moreover, why do dealers trade on the exchange, even though it is more costly than in the OTC protocols? We study both questions by modeling dealers’ protocol choice at the level of individual trades and rely on several proxies of information and search-and-bargaining frictions as explanatory variables. We find that both types of frictions are relevant for dealers’ protocol choice. More informed trades are more likely routed via the exchange. Similarly, more informed OTC transactions are often intermediated by a broker. Finally, we also show that proxies for search frictions influence dealers’ protocol choice. Combined, these results highlight the complementary roles that the three protocols provide.

We document that dealers’ individual trading decisions are also reflected in the informational content of aggregate order flows across protocols. Regressing bond returns on the lagged order flows from all three protocols, we show that the order flows from both the exchange and broker trading predict bond returns a few days out and that this effect is more pronounced for the exchange. We find no predictability for the flow from bilateral interdealer trades. Crucially, this is consistent with the observed pecking order and our finding that more informed trades are more likely routed via the exchange or a broker.

To assess the relative importance of search-and-bargaining and information frictions in determining transaction costs within protocols, we estimate a transaction-level regression of OTC discount on proxies of both types of frictions. Importantly, our results account for the potential endogeneity arising from the joint determination of the choice of trading protocol and pricing decisions. We find that more informed OTC trades receive significantly lower discounts. Moreover, dealers with more bargaining power receive higher discounts, suggesting that both types of frictions are quantitatively important drivers of OTC discount. Although our analysis is focused on the D2D segment, we also show that there is significant pass-through of OTC discount from the D2D to the dealer-to-customer (D2C) market segment.

Our paper contributes to several strands of the literature. There is a long-standing interest in the functioning of OTC markets. Theories of such markets focus on search-and-bargaining frictions to explain pricing differences (e.g., Duffie et al. 2005, 2007). The empirical literature has confirmed the relevance of these frictions: for example, with regard to dealer network structure (Li and Schürhoff 2019) and systematic changes in yield spreads (Friewald and Nagler 2019). Therefore, it is a widely held view that exchange markets feature lower transaction costs than OTC markets if considered separately (Edwards et al. 2007, Bessembinder et al. 2020).

This view has been challenged, however, by Glode and Opp (2020). These authors compare OTC and exchange markets in a model with endogenous information acquisition. They find that OTC markets can improve allocative efficiency by providing dealers incentives to acquire expertise. On the other hand, it can also reduce welfare when this expertise is exploited for adverse selection. In this case, competitive limit order markets tend to dominate.

Although Glode and Opp (2020) consider OTC and limit order markets separately, there are studies that analyze these protocols jointly in a hybrid setting, where information frictions are commonly at the center (e.g., Seppi 1990, Lee and Wang 2025).² A key feature of Seppi (1990) is that in OTC protocols, dealers know the identity of their counterparties, whereas market orders on the exchange are anonymous. In this model, equilibria arise where market participants always trade OTC when they are uninformed and at least partially on the exchange when they are informed. Lee and Wang (2025) extend this theory to investigate the welfare implications of restricting OTC trading in a hybrid market. In their model, dealers in the OTC protocol price discriminate between traders who are either likely informed or likely uninformed. A key testable hypothesis is that “the exchange spread is strictly wider than the OTC spread.”³ Consistent with this prediction, we find that OTC transaction costs for Bunds are lower than on the exchange. Because brokers provide a higher degree of anonymity than bilateral OTC trading, this prediction might also apply to our finding that transaction costs via brokers are higher than in the bilateral OTC protocol.

Some prior work relates information to order flows (e.g., Seppi 1990, Grossman 1992). In Grossman (1992), dealers in upstairs (OTC) and downstairs (limit order) markets receive information from their respective expressed order flows. In addition, dealers’ or brokers’ knowledge of unexpressed order flow in the upstairs OTC market allows them to provide better trading conditions over the counter than on the exchange. Relatedly, the model of Seppi (1990) predicts that uninformed traders trade only OTC, whereas informed traders either trade only on the exchange or use mixed strategies. As a result, the order flow from the exchange should be more informative than the OTC order flow. Consistently, we find that in the interdealer Bund market, aggregate order flows from the exchange and from broker trades each carry predictive information about future returns, whereas we do not find such a result for the aggregate order flow from bilateral OTC trades.⁴

Relevant information can also be about the identity of traders. Meling (2021) investigates the role of post-trade anonymity on market quality. He shows that the introduction of post-trade anonymity on the Oslo stock exchange increased trading volume and improved bid-ask spreads. This is in line with our results, which also point to the relevance of anonymity for trading decisions. Specifically, we find that dealers frequently forego some OTC discount to trade via brokers who provide anonymity.

Another key friction that has been identified in the literature is immediacy. Menkveld et al. (2017) observe that volatility shocks shift market shares between dark and lit venues along a pecking order of transaction costs. They present a simple model where the transaction cost differences are explained by varying degrees of immediacy. We complement their analysis by showing that the three trading protocols for which we document a pecking order of transaction cost differ not only in terms of immediacy but also, along the additional dimensions of transparency and anonymity. The exchange provides anonymity as well as the highest degree of immediacy and therefore, features higher transaction costs than the OTC protocols. Relative to bilateral OTC trading, brokers provide a higher degree of anonymity and potentially also immediacy, and thus, they have higher transaction costs, in line with theories based on information frictions. Indeed, we show that both exchange and broker order flows are informative about future returns.

Several papers have empirically studied pricing and protocol choice in hybrid markets in different asset classes. Barclay et al. (2006) analyze the choice between electronic and voice brokerage for U.S. Treasuries that go off-the-run, and Hendershott and Madhavan (2015) study U.S. corporate bonds trading both OTC and on a request-for-quote (RFQ) platform. More recently, Allen and Wittwer (2023) study traders’ venue choice in the Canadian sovereign bond market and analyze the welfare implications of modifications to the hybrid market setting. Investigating the market for index CDS (Credit Default Swap), Riggs et al. (2020) focus on the strategic choices of dealers and customers in trading via RFQ or request-for-streaming protocols, whereas Collin-Dufresne et al. (2020) report that in the interdealer segment, low-cost, low-immediacy protocols are preferred to trading on a limit order book. Studying the introduction of OTC trading to the Chinese interbank FX (foreign exchange) market, Holden et al. (2021) provide evidence for the role of bargaining power by comparing the aggregate price functions of small and large banks. A key advantage of our analysis is that we can directly compare the transaction costs for exchange and OTC trading as well as measure the relevant frictions at the transaction level. Relatedly, a number of papers have analyzed alternative protocols for equities, such as upstairs markets (e.g., Bessembinder and Venkataraman 2004).

We also contribute to the literature on the role of brokers as intermediaries. In the equity market, Barbon et al. (2019) and Di Maggio et al. (2019) document that brokers leak private information from informative trades to their institutional clients, whereas Han et al. (2024) argue that brokers help facilitate liquidity provision. In the U.S. Treasury market, Anderson and Liu (2024) show that brokerage platforms help dealers manage their interest rate risk. We document that brokers also play a key role in the Bund market as they represent the most actively used interdealer trading protocol. In particular, dealers benefit from the anonymity and facilitation of large orders provided by brokers. Brokers are not unique to the Bund market but are also an integral part of other interdealer fixed-income markets, such as U.S. Treasuries (Copeland et al. 2018), UK Gilts (Holland 2002), and Canadian sovereign bonds (Berger-Soucy et al. 2018). Our results thus highlight that understanding the role of brokers is crucial for modeling the microstructure of sovereign debt markets.

In sum, our findings point to a gap in the theoretical literature on OTC and exchange markets. A realistic description of the prevalent market structure should consider the coexistence of more than two different trading protocols and especially, the important role of interdealer brokers. Moreover, it should account for both information and search-and-bargaining frictions in a unified framework.

The remainder of this paper proceeds as follows. Section 2 describes the Bund market and the features of the three trading protocols as well as our data set. Section 3 introduces our main quantity of interest, OTC discount, and establishes a pecking order of transaction costs across protocols. In Section 4, we relate dealers’ protocol choice to proxies of information and search-and-bargaining frictions. We also show that the pecking order is consistent with the informational content of order flows across protocols. In Section 5, we study the role of both types of frictions in driving differences of OTC discount within trading protocols. Section 6 concludes.

2. The Bund Market

German sovereign debt securities enjoy benchmark status in the euro area and worldwide as a liquid and safe asset. Secondary market trading occurs among dealers and between dealers and customers. We refer to these as the interdealer (D2D) segment and the D2C segment of the market, respectively. In this paper, we focus on the D2D segment, which features a hybrid market structure with three distinct trading protocols: the interdealer exchange MTS, over-the-counter transactions intermediated by interdealer brokers, and bilateral OTC transactions. In this section, we describe the market structure of the three protocols and highlight their properties in terms of transparency and anonymity.⁵

2.1. Trading Protocols and Market Structure

2.1.1. Exchange.

The first option for Bund dealers to trade is via an exchange. The main exchange platform with a significant market share is the MTS interdealer exchange, which is operated as a fully electronic limit order book market. Dealers actively quote executable limit orders on MTS, and the depth on both the bid and ask sides of the book is typically in excess of 100 million Euro for most bonds, whereas the minimum trade size is 2.5 million Euro. Subscription services, such as Bloomberg, disseminate MTS prices and volumes at the best bid and ask levels in real time. Hence, this information is also available to non-MTS dealers. Similarly, all trading activity on the exchange is observed by all market participants in an anonymized way. Hence, MTS provides both pretrade transparency (about trading conditions) and post-trade transparency (about executed trades). After a trade occurs, the identity of the counterparties is revealed only to the two involved dealers for the purpose of settlement. This protocol, therefore, provides anonymity pretrade and post-trade from noninvolved parties.

2.1.2. Bilateral.

Dealers trade directly with one another in bilateral over-the-counter negotiations. This protocol represents the classical OTC trading protocol studied in most of the literature on OTC bond markets. Bilateral OTC trading is neither pretrade transparent nor post-trade transparent (trades are observed by the counterparties but not by other market participants). In these bilateral trades, both involved dealers are aware of the identity of their counterparty (i.e., bilateral trades are not anonymous for the counterparties involved). Other market participants do not observe any information about the trade.

2.1.3. Interdealer Brokers.

Dealers can also trade with one another via interdealer brokers. In this case, the initiating dealer communicates the trade request to a broker, who then undertakes to find a suitable counterparty and earns a fee. According to conversations with market participants, broker trades in the Bund market are still commonly done by voice. Just as bilateral OTC trades, broker trades are not transparent and unobserved by other market participants.⁶ Crucially, brokers follow a “matched principal” protocol. This implies that the broker acts as a counterparty to both sides of the trade without taking inventory and that the identities of counterparties are not revealed to each other, even after the trade is settled.⁷ This protocol, therefore, also provides anonymity both pre- and post-trade because the broker masks the identity of the other counterparty.

In what follows, we refer to bilateral and broker trades jointly as OTC trades in contrast to exchange trades on MTS.

Table 1 summarizes the transparency and anonymity features of the three protocols. The main differences between the three protocols are as follows. First, exchange trading is pre- and post-trade transparent, whereas OTC trades are not and thus, are unobserved by other market participants. Second, in bilateral trades, dealers are aware of each others’ identity, whereas the involved counterparties are anonymous in broker-intermediated trades. On the exchange, the identities of the counterparties are only revealed post-trade to the involved dealers for settlement purposes. Third, trading on the exchange is immediate, whereas both OTC protocols include search-and-bargaining processes.

Table 1. Comparison of Trading Characteristics Across Protocols

Importantly, all three trading protocols are simultaneously available to dealers, and in our sample, dealers rely on all of them. The fact that all three protocols are regularly used suggests that they offer complementary features to dealers for different trades. Of note, this hybrid structure is not unique to the interdealer Bund market but is similar to that of other European sovereign bond markets, such as those for sovereign debt issued by France and the United Kingdom. In the U.S. Treasury market (Fleming 1997, Fleming and Remolona 1999, Mizrach and Neely 2009), the order book of the BrokerTec platform assumes a role similar to that of MTS in the euro area. Interestingly, this market structure differs from that studied in most theoretical models, which to the best of our knowledge, do not consider all three protocols jointly.

2.2. Data

We focus on German sovereign bonds with maturities of 2, 5, 10, and 30 years, which we jointly refer to as Bunds. Reissuances are common for Bunds, and on-the-run effects are much less important than for U.S. Treasuries. We, therefore, include both on-the-run and off-the-run bonds in our analysis.

Our study is based on a unique regulatory transactions data set of all Bund trades involving at least one German financial institution from June 2011 to December 2017. Importantly, this trade repository includes interdealer trades from all three market segments in addition to bilateral D2C trades. The transactions data are based on reporting requirements of German financial institutions to the German Federal Financial Supervisory Authority (Bundesanstalt für Finanzdienstleistungsaufsicht, popularly known as “BaFin”). They contain information on the price, size, time, and direction of each trade as well as a flag indicating whether a trade was over the counter or the platform in which the trade was executed. The data set also includes anonymous identifiers for the reporting agent and the counterparty of a trade. We match these data with the full limit order book data from MTS.

Table 2 describes the trading activity covered by our sample and the subsamples relevant for our analysis in terms of number of trades, aggregate trading activity, trade size, and market share by type of trading protocol. Our full sample contains over 500,000 trades across 116 German federal bonds and 402 reporting institutions, corresponding to a total nominal volume of almost 3.4 trillion Euro (labeled “full sample”). A survey by the German finance agency among Bund Issues Auction Group members pegs daily trading volume in the secondary market at more than 17 billion Euro. Our sample captures about 11% of this trading activity.

Table 2. Trading Activity by Subsamples

For about 210,000 of these trades (corresponding to a volume of roughly 1.5 trillion Euro), we are able to identify both initiator and counterparty (labeled “Init. & counterp. ID known” in Table 2).⁸ We classify the involved parties as either dealers or customers based on their access to the interdealer exchange MTS. According to this classification, we observe about 47,500 interdealer trades between MTS dealers (labeled “D2D”). For trades on its platform, MTS imposes a 2.5 million Euro minimum trade size. To ensure comparability with the OTC segment of the market, we limit our sample to the set of trades above that threshold. These roughly 22,000 interdealer trades (by 34 dealers and via six interdealer brokers) represent more than 95% of the overall interdealer trading volume. About 2,200 interdealer trades take place on MTS (labeled “D2D via MTS,” corresponding to 3.7% of overall observed interdealer volume), whereas there are about 6,800 OTC transactions resulting from bilateral negotiations (“D2D via bilateral OTC,” corresponding to 18.4% of interdealer volume). The by far largest share of interdealer trading is done via brokers (“D2D via broker”) in terms of both the number of trades (about 13,300) and overall volume (77.9%).⁹ The average nominal trade size in the interdealer segment is lowest on MTS (7.9 million Euro), somewhat larger (12.7 million Euro) in bilateral OTC transactions, and considerably larger (27.6 million Euro) in broker-intermediated trades.

3. OTC Discount in the Bund Interdealer Market

In this section, we document that transaction costs in the interdealer Bund market are on average lower for OTC trades compared with contemporaneous quotes on the exchange. Moreover, bilateral OTC trades are cheaper than those intermediated via brokers. In Section 3.1, we define our measure of transaction cost differential, which we label “OTC discount.” In Section 3.2, we provide descriptive statistics about the transaction cost measure across trading protocols. Section 3.3 then confirms this result based on a matched sample of bilateral and broker-intermediated trades.

3.1. Definition

We measure the transaction cost of OTC trades relative to those on the exchange protocol of the interdealer Bund market. In doing so, we make use of the full limit order book of MTS and compare the transaction costs of OTC trades with the contemporaneous trading conditions on the exchange. Our quantity of interest is the premium or discount that OTC trades pay or receive with respect to bid and ask quotes on MTS when initiating a trade. We define the OTC discount as the difference between the price that a trade would have incurred on the exchange and the observed price of an OTC trade. We measure OTC discount as a fraction of the hypothetical transaction cost on the exchange. Formally, OTC discount is defined as

where $\mathit{\text{pric}}{e}^{\text{MTS}}$ is a reference price that the same trade would have incurred on MTS at the same time.¹⁰ The denominator in Equation (1) is equal to the quoted half-spread on MTS.

Figure 1 illustrates the trade sign identification and calculation of OTC discount. A positive OTC discount implies that executing a trade over the counter is cheaper for the initiator than trading on the exchange. Trades at the midprice correspond to 100% OTC discount, the upper bound, whereas buys at the MTS ask price and sells at the MTS bid price have 0% OTC discount; that is, the OTC trade occurred at the same price as on the exchange. When OTC discount is negative (i.e., when it would have been cheaper to trade on MTS instead of OTC), we refer to this as an OTC premium.

As the reference price, we choose the quoted price at the respective best level of the limit order book (i.e., $\mathit{\text{pric}}{e}^{\text{MTS}}$ is the best ask (bid) price for buyer-initiated (seller-initiated) trades). Hence, we disregard that larger trades would also have to execute against quotes at deeper levels of the limit order book (known as “walking up the book”). We do so for several reasons. First, large trades are rare on MTS, and the price at the best provides a better benchmark for OTC trades than an “effective” price. Second, in choosing the quoted price at the best as $\mathit{\text{pric}}{e}^{\text{MTS}}$, our measurement for OTC discount is conservative (i.e., smaller) with respect to a price that takes into account the depth of the order book.

Given that the market share of exchange transactions is substantially lower than that of OTC trading, one might worry about the use of exchange quotes as a reference. However, market share is endogenous to trading conditions, and all else equal, dealers will generally gravitate to cheaper protocols. A better indicator of the potential capacity of the exchange to facilitate trading is the depth of the order book. We find that 98% of all OTC transactions could take place on the MTS order book at the moment of the trade. Moreover, notwithstanding the smaller market share, all limit orders on MTS are executable. MTS bid and ask quotes are thus an appropriate reference price.

3.2. Properties of OTC Discount

Table 3 shows the descriptive statistics of OTC discount for the same subsamples of OTC trades as those discussed above. Both mean and median of OTC discount are positive across all considered subsamples, and we reject the null hypothesis that the average OTC discount is zero at the 0.01% significance level using a simple t-test. For interdealer trades with a size of at least 2.5 million Euro (“trade size $\ge$2.5 million Euro”), the average OTC discount equals 34.7%. This implies that dealers pay on average just about two thirds of the quoted MTS bid-ask spread when trading among each other over the counter. The median OTC discount is even larger at 61.6%, and the trade at the 95th percentile receives a discount of 96%. Less than 16% of D2D trades above 2.5 million Euro incur an OTC premium.

Table 3. Descriptive Statistics of OTC Discount

The existence of an OTC discount implies that it is typically cheaper for dealers to execute Bund transactions in the over-the-counter market. The difference in transaction costs between OTC and exchange trades is not only highly statistically significant but also, economically meaningful. A simple back-of-the-envelope calculation suggests that an OTC discount of 35% implies a daily cost advantage of about 2.4 million Euro (600 million Euro annually) from trading OTC rather than on the exchange.¹¹

In Online Appendix C, we show that these results are robust with respect to alternative definitions of OTC discount. First, we explicitly account for the effects of large trades walking up the book (i.e., $\mathit{\text{pric}}{e}^{\text{MTS}}$ is no longer the price at the best but the actual price that would result for a trade of a given size given the current depth of the limit order book). Second, in order to relate to yields instead of transaction costs, we consider an alternative definition of OTC discount where we normalize by price instead of transaction cost (i.e., the denominator in Equation (1) becomes $\mathit{\text{pric}}{e}^{\text{mid}}$). Both alternative definitions equally give rise to a pervasive and economically large OTC discount.

3.3. Difference in OTC Discount Across Protocols

The results documented in the previous section mask sizable differences in the distribution of OTC discount between bilateral and broker OTC trades. As shown in Table 3, bilateral OTC trades (“D2D via bilateral OTC”) receive a considerably larger average discount of 54.6% as opposed to 24.6% for brokered trades (“D2D via broker”). Bilateral trades also have a lower share of trades with an OTC premium: 6.9% compared with 20.4% for trades via brokers.

Figure 2 shows the histogram of OTC discount for interdealer trades with a minimum size of at least 2.5 million Euro, where panel (a) refers to bilaterally negotiated trades and panel (b) is based on transactions via brokers. Both subsets feature a distribution that is heavily tilted toward positive values of OTC discount (i.e., in the majority of cases, trading over the counter is cheaper than on the exchange). Moreover, bilateral OTC trades clearly feature more mass on large positive values of OTC discount than broker trades. In summary, a pecking order of average transaction costs emerges; bilateral is cheaper than broker-based OTC trading, and both are cheaper than transactions on the exchange. This is surprising given that 78% of the transaction volume in the D2D segment is executed via brokers, implying that dealers prefer to route the majority of their transactions through a more expensive protocol.

Of course, the OTC discount differential between bilateral and broker-intermediated trades could be because of systematic differences in trade characteristics between the two protocols. For example, as documented above, broker trades are often larger in volume than bilateral ones. To fully establish the pecking order of transaction costs, we consider similar trades that were carried out via different trading protocols. Specifically, we construct a matched sample that pairs bilaterally negotiated trades with similar broker transactions using nearest neighbor propensity score matching.¹² We then estimate the following equation:

where ${𝟙}_{\mathit{\text{broker}}}$ indicates a dummy variable that assumes the value of one for broker trades and where ${\mathit{v}}_n$ denotes a rich vector of trade-level characteristics. We further include both bond and dealer fixed effects.

Table 4 shows the results. The main finding is that broker-facilitated trades feature an OTC discount that is about 20 percentage points lower than for bilaterally negotiated trades in the matched sample. Although this is somewhat smaller than the difference shown in the summary statistics above, it highlights that broker trades receive a substantially lower OTC discount than similar bilaterally negotiated trades, controlling for a host of potential confounding factors.

Table 4. Differences in OTC Discount Across Interdealer Protocols

4. Information, Protocol Choice, and OTC Discount

Our analysis thus far has established a pecking order of transaction costs in the interdealer Bund market. Trades on the exchange are more expensive than OTC trades. Among OTC trades, bilateral trades have lower transaction costs than broker-intermediated OTC trades. These findings are difficult to rationalize with theories relying on search-and-bargaining frictions in OTC markets. First, if such frictions were the main driver of transaction costs, then the exchange protocol, where they are absent, should be the cheapest. Second, as search costs are arguably lower for broker-intermediated trades than bilateral OTC trades, they should feature a higher, not a lower, discount.

Both of these findings can be explained, however, by information motives. Liquidity providers on the exchange might charge a wider spread to protect themselves from being adversely selected by more informed traders. Similarly, dealers providing liquidity via brokers may request wider spreads because the broker protocol conceals the identity of the counterparty, who may be more informed. This section provides empirical evidence supporting this view.

We start by documenting that dealers’ choice of trading protocol is often driven by variables measuring the information content of a given trade. We then show that the pecking order of transaction costs is aligned with the informational content of order flows across protocols for predicting Bund returns.

4.1. Protocol Choice in the Interdealer Segment

In the interdealer Bund market, dealers face the choice between three trading protocols: exchange, bilateral, or via a broker. We model this choice as a sequential decision. First, we consider the choice between trading on the exchange or OTC.¹³ To this end, we estimate a probit model where the dependent variable equals one for over-the-counter trades and zero for trades on MTS. Conditional on a trade not being executed on the exchange, we then estimate a second probit model where the dependent variable equals one for trades via a broker and zero otherwise (i.e., bilaterally negotiated OTC transactions). In both models, the probability to route a trade via a specific protocol depends on a set of proxies for information and search-and-bargaining frictions that we detail next.¹⁴

4.1.1. Measuring Information and Search-and-Bargaining Frictions.

We now motivate our choice of regressors to capture the potential drivers of protocol choice. Variable definitions take the point of view of the trader initiating the transaction.

4.1.1.1. Information Frictions.

In Seppi (1990), equilibria emerge where liquidity traders prefer to trade large blocks over the counter, whereas informed traders may also split up an order into a series of smaller trades. We create a dummy variable, Order Splitting, that takes the value of one when a trade is the result of order splitting and zero otherwise. We identify order splitting when a dealer has multiple trades of the same bond in the same direction (buy or sell) with at most 48 hours between two transactions and without any offsetting trades.

Our second variable measuring information is based on the time elapsed since a dealer last traded the same security and is similar in spirit to the one used by Brancaccio et al. (2020). The idea is that dealers have more current information about securities that they have traded recently. More precisely, Informedness is defined as the negative of the logarithm of the time since a dealer last traded the same bond.¹⁵

Information is often revealed via order flow. Czech et al. (2021) show that the order flow by hedge funds and mutual funds predicts bond returns in the UK Gilt market, and Ranaldo and Somogyi (2021) present similar findings for the FX market. These results are consistent with Grossman (1992), in which information need not necessarily be about fundamental value but can also refer to expressed and unexpressed order flow. We use two proxy variables that indicate the direction of orders in the market. Aggregate Order Flow captures for each trade the net order flow of all preceding OTC trades on the same day, including customer trades, in all Bunds. Order Book Imbalance measures the contemporaneous imbalance between the best three levels on both sides of the limit order book of the same Bund. Both variables are defined as positive for trades in the direction of the market. We associate trading in the same direction as the market with a higher degree of information.

In line with information-based theories of hybrid markets, our hypothesis is that more informed transactions are routed via the exchange instead of OTC. Moreover, because the broker protocol provides anonymity, we expect that among all OTC transactions, trades with higher information content are more likely to be broker intermediated.

4.1.1.2. Search-and-Bargaining Frictions.

Search-based models of over-the-counter markets identify bargaining power as the crucial driver of OTC transaction costs. They predict that dealers with more bargaining power face lower transaction costs (Duffie et al. 2005, 2007; Duffie 2012). According to these theories, dealers will transact in the cheapest protocol. On the exchange, dealers do not face search costs and cannot exercise bargaining power.

Dealers’ inventory is a key factor in trading and pricing decisions (e.g., Stoll 1978, Amihud and Mendelson 1980). Inventory requires balance sheet capacity (Dick-Nielsen and Rossi 2019, Goldstein and Hotchkiss 2020, Colliard et al. 2021). Dealers with large inventory are more likely to be constrained and hence, in a worse bargaining position. We measure a dealer’s Inventory as her net imbalance over all Bunds on the same day prior to the trade divided by her average daily trading volume. The variable is signed so that it is positive for trades increasing a dealer’s net inventory position in absolute value. Search costs amplify inventory risk on volatile days, which reduces dealers’ bargaining power. We measure Volatility as the intraday price volatility of each bond on MTS. In line with the theories based on search and bargaining, we expect the probability to trade OTC to depend negatively on both Inventory and Volatility.

Previous studies have identified the dealer network structure and trading relations as important factors for pricing in OTC markets (e.g., Di Maggio et al. 2017, Li and Schürhoff 2019, Hendershott et al. 2020). Unfortunately, we only observe trades involving German financial institutions and are thus unable to reconstruct the full network structure of interdealer trades. Therefore, we use each dealer’s overall trading volume (Dealer Volume) as an indirect measure of network centrality (cf. Nikolova et al. 2020) and by extension, bargaining power. Our hypothesis is that dealers with a higher overall trading volume are more likely to trade OTC.

Duffie et al. (2005) argue that access to outside options increases dealers’ bargaining power. In the Bund market, dealers have the outside option to trade on the exchange. Because larger trades become increasingly expensive on MTS, this outside option vanishes with increasing Trade Size. Therefore, we expect larger trades to be less likely done via MTS. Moreover, as search costs are likely higher for larger trades and as brokers alleviate search costs, the probability to route an OTC trade via a broker should also increase in trade size.

Online Appendix E provides detailed definitions of all proxy variables as well as control variables in Table A.8 and their descriptive statistics in Table A.9.

4.1.2. Choice Between Exchange and OTC.

We study dealers’ choice between using the exchange or going OTC via the following probit model:

where $\mathrm{\Phi }$ is the standard normal cumulative distribution function and ${\mathit{\omega }}_{I,n}$ is a vector of trade-level proxies for search-and-bargaining and information frictions as well as control variables.

To account for potential nonlinearities with respect to trade size, we employ dummy variables for trades of 10–30 million Euro and for trades larger than 30 million Euro. The baseline hence consists of trades for 2.5–10 million Euro. In addition, Round Trade Size is a dummy variable that equals one for trades with a nominal amount of exactly 2.5, 5, or 10 million Euro, which are typical trade sizes on MTS, and zero otherwise. Controls is a vector of further control variables that capture bond-specific liquidity and bond characteristics.¹⁶ Throughout our analysis, all regressors except for dummy variables have been standardized to have a mean of zero and unit variance.

Specifically, we have

Table 5 provides the regression results. The first three columns in Table 5 show the marginal effects of the probit estimation of Equation (3), with standard errors clustered at the dealer level. We first analyze the role of information frictions in driving protocol choice. Our two measures of private information, order splitting and informedness, point in the same direction; more informed trades are more likely to be routed via MTS. Transactions that have been split into smaller trades are about two percentage points more likely to trade on MTS. Moreover, the probability to trade OTC increases by 0.75–1 percentage point with each standard deviation increase in informedness. In contrast to our expectation, we find weakly significant evidence that aggregate market-wide information (aggregate order flow and order book imbalance) increases the probability to trade OTC.

Table 5. Probability Model for Protocol Choice

We now turn to the role of search-and-bargaining frictions as drivers of protocol choice. Dealers holding larger inventory of a given security are more likely to transact them OTC. A one-standard-deviation increase in inventory of a given security translates into a 0.66- to 0.85-percentage-point-higher probability that the transaction is OTC. On more volatile days, trading on the exchange is more likely. A one-standard-deviation increase in volatility increases the likelihood of routing a trade to the exchange by about three percentage points, in line with the notion that the outside option of search in the OTC market is less feasible.

Finally, larger transactions are less likely to be traded on the exchange. Trades in the 10–30 million Euro range are 1.9–2.6 percentage points more likely to be traded over the counter with respect to the baseline of trades for 2.5–10 million Euro. This effect is even more pronounced for trades larger than 30 million Euro, where OTC trading is about 8.7–10.5 percentage points more likely. Round trade sizes are significantly more likely on MTS.

In summary, these results show that both information and search-and-bargaining frictions impact dealers’ decision to trade on the exchange or in the OTC market. We next study the role of these frictions for the decision to trade bilaterally or via a broker.

4.1.3. Choice Between Broker and Bilateral OTC.

More than 75% of all interdealer trading is done via brokers. We model dealers’ decision to route an OTC trade via a broker or trade bilaterally with the following probit model:

where ${\mathit{\omega }}_{II,n}$ captures essentially the same set of trade-level proxies for information and search-and-bargaining frictions as well as trade-level controls. The only difference with respect to the set of explanatory variables ${\mathit{\omega }}_{I,n}$ used before is that we do not include a dummy for round trade sizes as these are much less prevalent for OTC transactions.

The last three columns in Table 5 show the marginal effects of this probit model. We highlight the following results. First, more informed transactions are more likely to be intermediated by a broker than bilaterally negotiated; transactions that have been split into smaller trades are more than six percentage points more likely to be routed via a broker. That said, the alternative measure of informedness does not significantly affect the decision to trade via a broker.

Second, dealers with larger inventories are more likely to trade via brokers. A one-standard-deviation increase in dealer inventory increases the likelihood of broker intermediation by about two percentage points. Third, large dealers are more likely to trade via brokers than smaller ones. A dealer with a one-standard-deviation-larger total trading volume is about 14% more likely to trade via a broker. Fourth, larger trades are significantly more likely to be routed via brokers. OTC trades in the 10–30 million Euro range are more than 20 percentage points more likely to be broker intermediated. Trades larger than 30 million Euro are almost 40 percentage points more likely to be routed via a broker.

Our interpretation of these results is as follows. Larger and more central dealers as well as dealers with larger trading needs avoid trading bilaterally and instead, prefer to trade anonymously via brokers. Possible motives could be to maintain their informational advantage (Holland 2002, Babus and Kondor 2018) or to protect themselves against front-running (Harris 1997, Brunnermeier and Pedersen 2005) or back-running (Yang and Zhu 2020).¹⁷

In sum, our results suggest that more informed transactions are more likely to take place on the exchange. Among OTC trades, more informed transactions are more likely to be routed via an interdealer broker. Combined, they highlight the important role of information frictions in dealers’ trading decisions.

4.2. The Informational Content of Order Flows

In the previous section, we have shown that variables measuring the informational content of trades are strong predictors of the protocol in which dealers execute transactions. If there are systematic differences in the informational content of the three trading protocols, these should be reflected in aggregate order flows. Moreover, information is often associated with realized returns. We, therefore, follow the literature on informed trading in bond markets (e.g., Czech et al. 2021) to assess whether the informational content of trades is in line with the pecking order of transaction costs that we have documented. Specifically, we regress bond returns on order flows in each protocol as follows:

Here, ${\mathit{\text{return}}}_{i,d,d+\mathrm{\Delta }}$ denotes the cumulative log return of bond i from day d to day $d+\mathrm{\Delta }$ in basis points. ${\mathit{\text{MTS}} \mathit{\text{order}} \mathit{\text{flow}}}_{i,d}$, ${\mathit{\text{Broker}} \mathit{\text{order}} \mathit{\text{flow}}}_{i,d}$, and ${\mathit{\text{Bilateral}} \mathit{\text{order}} \mathit{\text{flow}}}_{i,d}$ are the net total volume for each bond and trading day.¹⁸ Control variables include the customer order flows, the lagged return, the logarithm of the amount outstanding in the bond, and the logarithm of the remaining maturity. We also include bond and day fixed effects.

Table 6 shows the results of this regression for horizons $\mathrm{\Delta }$ equal to one, two, and five days. MTS order flow is statistically significant for the two- and five-day horizons and has the strongest price impact at all horizons. The coefficient on order flow via brokers is slightly smaller and statistically significant at the 5% level for the two-day horizon. By contrast, bilateral order flow is indistinguishable from zero at all horizons.

Table 6. Bond Returns and Interdealer Order Flows

Our findings on the predictive content of order flow are consistent with the pecking order of transaction costs documented in Section 3. Trading on MTS is the most informative about future returns and also features the highest transaction cost. Trading via brokers has somewhat less predictive power and lower transaction costs. Bilateral trading, in turn, carries no information about future returns and features the lowest transaction costs. These results are also in line with our previous findings that more informed transactions are more likely done on the exchange and that better informed OTC transactions are more likely to be broker intermediated. Thus, information frictions represent a coherent explanation for the observed differences in transaction costs and dealers’ protocol choice.

5. Transaction Cost Differentials Within Protocols

In the previous sections, we have established a pecking order of transaction costs, which is consistent with informational differences across trading protocols. We now turn to an analysis of the drivers of OTC discount within protocols. To this end, we regress OTC discount on proxies of information and search-and-bargaining frictions.

5.1. Regression Specification

Transaction costs likely influence dealers’ protocol choice. Hence, trading protocol and OTC discount are likely determined jointly. To correct for this potential endogeneity in modeling OTC discount, we follow the standard approach of using a two-stage switching model (Madhavan and Cheng 1997, Bessembinder and Venkataraman 2004, Hendershott and Madhavan 2015). The first stage consists of estimating a model of protocol choice as in Section 4.1. In the second stage, we then relate OTC discount to explanatory variables and the inverse Mills ratio calculated from the first stage.

Assuming that error terms are jointly normal, we estimate the following equation separately for bilateral and broker OTC trades at the level of individual trades (indexed by n):

where $s\in \{\text{bilateral},\text{broker}\}$ indicates the trading protocol used and where v is a vector collecting proxies for information and search-and-bargaining frictions as well as a set of control variables. As selectivity adjustments, we include the inverse Mills ratio ${\widehat{\lambda }}^{\text{OTC}}$ to control for the decision to trade over the counter and ${\widehat{\lambda }}^{\text{bilateral}}$ or ${\widehat{\lambda }}^{\text{broker}}$ to control for the decision to trade bilaterally or via a broker, respectively. The explanatory variables in Equation (7) arewhere ${\mathrm{\Delta }}_b$ and ${\mathrm{\Delta }}_i$ are bond and initiating dealer fixed effects, respectively, and Controls is a vector of further control variables as above.¹⁹ As an additional proxy for information frictions, we now also include Price impact as the price change 15 minutes after each trade following Collin-Dufresne et al. (2020).²⁰ We interpret trades with larger price impact as being more informed.

5.2. Estimation Results

Table 7 shows the estimation results, with standard errors obtained via bootstrap to account for correlation with protocol choice.²¹

Table 7. Intermediation Frictions and OTC Discount

5.2.1. Bilaterally Negotiated Interdealer Trading.

Column (1) in Table 7 reports the coefficient estimates for bilaterally negotiated OTC trades. A one-standard-deviation increase in price impact is associated with a 2.2-percentage-point-lower OTC discount. Order splitting has a similarly large effect, which is not, however, statistically significant.²² We also do not find a statistically significant impact of informedness in the presence of the above two proxies for private information. The coefficient for aggregate order flow is negative and highly significant, implying that trades in the same direction as the overall market feature a lower OTC discount. These results suggest that dealers, at least to some extent, price discriminate against each other based on their counterparty’s perceived information.

We next turn to the role of search-and-bargaining frictions. Volatility, which makes the outside option of continued search more costly or less feasible, strongly raises transaction costs. A one-standard-deviation increase in volatility reduces OTC discount by about 10 percentage points. OTC discount also decreases with trade size. A one-standard-deviation-larger trade is associated with a 4.2-percentage-point-lower OTC discount.²³ Importantly, our regression includes dealer fixed effects, which absorb any residual time-invariant differences in bargaining power across dealers. The estimated fixed effects show sizeable variation; the standard deviation across dealers amounts to 17.6 percentage points of the MTS half-spread—almost one third of the average OTC discount.

To further illustrate the effect of dealer-specific bargaining power, we construct dummy variables for the top and bottom five dealers in terms of their dealer fixed effects in column (1) in Table 7. In column (2) in Table 7, we repeat the estimation, including these dummies instead of dealer fixed effects. The top five dealers receive a 7.5-percentage-point-higher OTC discount, whereas the bottom five obtain a 18.3-percentage-point-lower OTC discount relative to their peers. These effects are again sizeable in light of the average OTC discount of 54.6%. This suggests that dealer-specific bargaining power plays a quantitatively important role for transaction cost over and above the proxies included in our regressions.

We show the robustness of our results in Online Appendix C. We consider alternative definitions of OTC discount, where we (a) normalize OTC discount by price instead of transaction cost and (b) include the effect of larger trades walking up the limit order book and consuming liquidity from deeper levels of the limit order book. All results are quantitatively and qualitatively in line with those presented here.

5.2.2. Broker-Intermediated Interdealer Trading.

We now investigate the determinants of OTC discount for trades intermediated by interdealer brokers. Column (3) in Table 7 shows the results. They highlight that the same set of drivers as for bilateral OTC trades affects transaction costs in the broker protocol.

Specifically, we find that higher price impact and the presence of order splitting are both associated with significantly lower OTC discounts. Although the coefficient on price impact is somewhat smaller compared with bilateral trades, the order splitting dummy is highly significant and almost twice as large. Trades that are part of a meta-order receive a discount that is 4.2 percentage points lower. Although the coefficient on informedness is negative as expected, it is not statistically significant. Trading in the same direction as the market also significantly reduces OTC discount. A one-standard-deviation-higher aggregate order flow implies a 1.5-percentage-point-lower OTC discount for broker-facilitated trades.

We now turn to measures of search-and-bargaining frictions. As before, days with higher volatility are characterized by substantially lower OTC discounts. The impact of volatility is about 50% larger than for bilateral trades and highly statistically significant. As for bilateral transactions, larger trades receive significantly smaller discounts. A one-standard-deviation increase in log trade size is associated with a 3.2-percentage-point-lower OTC discount.

Combined, these results show that more informed trades receive lower OTC discounts, in line with information-based theories of hybrid markets. At the same time, we find that less bargaining power translates into lower OTC discounts, consistent with search-and-bargaining-based theories of OTC markets. Although the coefficients differ slightly, the role of information and search-and-bargaining frictions is comparable across the two OTC protocols.

5.3. The Dealer-to-Customer Segment

Although our paper focuses on the hybrid market structure of the Bund interdealer segments and their main drivers, an interesting complementary question is how this structure and dealers’ pricing and trading decisions in the interdealer segment affect trading conditions in the dealer-to-customer segment. In Online Appendix F, we extend our analysis to the D2C segment of the Bund market and present three main findings. First, we observe an OTC discount with respect to MTS quotes even for D2C trades, suggesting that dealers pass on a substantial share of their transaction cost advantage to their clients. Second, information and search-and-bargaining frictions play a similar role in driving OTC discount as in the D2D segment. Third, dealers with more bargaining power in the interdealer segment offer lower transaction costs to their clients.

6. Conclusion

In an environment where academics and regulators increasingly call for a shift from traditional over-the-counter market structures toward electronic platforms and greater transparency, understanding the drivers of dealers’ pricing decisions across trading protocols is ever more important. Using a unique regulatory data set of securities transactions, our paper contributes to this debate along several dimensions. We find that the vast majority of OTC trades execute at favorable prices relative to the exchange limit order book. This OTC discount is significantly larger for bilateral than for broker-intermediated OTC trades, giving rise to a pecking order of transaction costs. Dealers’ decisions of where to execute a trade depend both on their bargaining power as well as on the information that they have. Consistently, we document that this pecking order is in line with the relative informedness of order flows from the different protocols. Moreover, we show that the variation of OTC discount across transactions within the same protocol is driven by both information and search-and-bargaining frictions, suggesting that both types of frictions are present in hybrid markets.

Our findings are relevant for the growing literature on OTC markets. Specifically, they suggest that both theoretical work and empirical work in this area should strive to account for the complex structure of today’s markets by considering (i) the coexistence of multiple trading protocols, (ii) the relevance of interdealer brokers, (iii) the importance of both information and search-and-bargaining frictions, and (iv) the interlinkages between the interdealer and D2C segments.

On the regulatory side, there is a strong effort to improve OTC market functioning. In Europe, the Markets in Financial Instruments Directive II (MiFID II) regulation was rolled out with the intention to improve market transparency, whereas in the United States, the Financial Industry Regulatory Authority (FINRA) has started to collect transaction-level data for sovereign bonds similar to that in the Trade Reporting and Compliance Engine (TRACE) database and the regulatory data set used in this paper. In recent years, there has also been a regulatory push away from OTC toward exchange-based trading. However, as pointed out by Dugast et al. (2022), a centralized exchange market is not always socially optimal as agents’ trading decisions are driven by their private incentives. Although our empirical analysis cannot shed light on the socially optimal market structure, our results suggest that there is substantial heterogeneity across trades and dealers. As a result, bilateral, broker-intermediated, and exchange-based trading protocols all play complementary roles in serving the different needs of dealers in a way that a single venue might not be able to achieve.