The Spread That Got Narrower—and the Friction That Did Not Disappear
The subject of this series is the evolving U.S. corporate-bond market: a dealer-intermediated over-the-counter market in which customers search for trades and dealers decide how much inventory and balance-sheet capacity to commit. The immediate question is why bid-ask spreads can become smaller without a comparable disappearance of the liquidity cost embedded in bond yields. The practical judgment this supports is simple but important: a lower trading cost is evidence about one part of liquidity, not a complete verdict on how easy or safe it is to move capital through the market.
A better question than “Did liquidity improve?”
Corporate bonds are a major source of long-term borrowing for large American companies. The market is large—outstanding corporate debt exceeded $11 trillion and average daily trading volume was nearly $60 billion in October 2025, according to figures reported by the paper—but it does not work like a stock exchange with one central order book. A customer who wants to buy or sell usually needs a dealer to locate the bond, quote a price, and, in some cases, hold the security long enough to complete the other side of the trade.
That arrangement makes the word liquidity do too much work. One meaning concerns the price of an immediate transaction. If a customer can buy at a price close to the level at which another customer could sell, the market has a narrow bid-ask spread. Another meaning concerns the cost of holding an asset when the owner may need to trade later. A bond can be cheap to trade at a particular moment and still require investors to accept compensation for the possibility that a future trade will be slow, difficult, or expensive.
The Philadelphia Fed working paper The Evolution of the Corporate Bond Market: A Theoretical Analysis makes this distinction explicit. Its model follows a representative bond in a dealer-intermediated market and separates the bid-ask spread from a liquidity yield spread. In the model, the bond is priced at par in the competitive inter-dealer market. The additional yield is therefore not meant to measure default risk or compensation for bearing ordinary price risk. It represents the liquidity cost created by search frictions, customer trading needs, and dealer balance-sheet constraints.
Those two measures can move together, but they need not. A bid-ask spread describes the concession made in one exchange. A liquidity yield spread describes the compensation required over the holding period because the owner may not be able to exchange the bond immediately when preferences change. The first is visible in the quote; the second can remain in the return investors demand even when quotes look more competitive.
The paper's combined numerical experiment makes the distinction concrete. It changes three features of the market at once: balance-sheet costs rise by 32 basis points, the electronic share of trading rises from 20% to 50%, and turnover rises from 11% to 16%. In the table's combined result, the liquidity yield spread increases by 0.87 basis points, or about 2.8%, while the average bid-ask spread falls by 6.29 basis points for voice trades and 4.41 basis points for electronic trades. Those are outputs of a calibrated model, not a direct forecast of every corporate bond. But the direction of the result answers the central puzzle: the market can look cheaper to trade while the modeled cost of carrying liquidity risk is almost unchanged.
The distinction matters to several groups at once. An investor looking only at a narrower spread may underestimate the compensation still required to hold a bond that could become difficult to sell. An issuer may see a better execution environment without receiving the full benefit of lower secondary-market frictions in the yield investors require. A dealer may face lower revenue per trade but more or less pressure on its scarce balance sheet depending on which side of the market customers are using. A single liquidity label cannot tell these stories apart.
Three changes, three routes through the same market
The market has not changed along one dimension. The paper studies three developments that operate through different channels.
First, post–Global Financial Crisis regulation increased the cost for many bank-affiliated dealers of carrying assets on their balance sheets. Capital and liquidity requirements, along with greater scrutiny of inventory holdings, made balance-sheet space a more expensive input to market making. The paper reports that the share of outstanding corporate and related bonds held by security brokers and dealers fell from roughly 2–3% in 2006 to less than 1% in 2018, while also noting that this measure includes more than corporate bonds alone. The broader point is not that one statistic proves a single regulatory effect. It is that the capacity available for dealers to absorb customer trades became economically important.
Second, more trading moved to electronic request-for-quote platforms. Instead of contacting one dealer by phone or chat, a customer can ask several dealers for prices at once. For investment-grade bonds, the paper reports that the electronic share grew from about 20% in 2015 to about 50% in 2025. That change directly affects search and competition: a customer has a better chance of finding a usable quote, and a dealer faces more pressure from competing offers. The result can be a lower transaction cost even if the underlying need for dealers to hold inventory remains.
Electronic access is not the same as universal electronic liquidity. Voice trading remains important for large blocks and lower-rated issues. A platform can contact several dealers, but a dealer that does not hold the bond may not be able to respond to a customer who wants to buy. The number of names on an RFQ is therefore not identical to the number of effective competitors. Competition depends on who has the right inventory at that moment.
Third, the owners of corporate bonds have changed. From 2000 through the second quarter of 2025, the share of corporate and foreign bonds held by mutual funds and ETFs rose from 9.5% to 23%, while the share held by life insurers and pension funds fell from 41% to 32%, according to the figures cited in the paper. The model does not treat every institutional difference separately. Instead, it represents the shift toward more active vehicles as more frequent changes in customers' desire to own the bond. A portfolio that responds to flows, benchmarks, index changes, or redemptions can create more occasions on which someone needs to buy or sell.
These developments should not be treated as three synonyms for modernization. Regulation changes the cost of carrying inventory. Electronic RFQs change how customers search and how dealers compete. More active ownership changes how often the market is asked to provide immediacy. Each can change a different part of the dealer's economic problem, so each can leave a different trace in quotes, yields, order flow, and inventory.
The burden can move even when the headline spread does not
The paper's central mechanism begins with a deliberately simple dealer choice. A dealer that holds the bond can sell it to a customer who wants to buy, but cannot buy another unit under the model's zero-or-one inventory assumption. A dealer that does not hold the bond has balance-sheet space and can buy from a customer who wants to sell, but cannot sell the bond. The assumption is not a literal description of every real position. It isolates a directional fact: inventory makes a dealer able to serve one side of customer demand, while unused capacity makes the other side possible.
That choice creates a hidden connection between individual trades and market-wide outcomes. If fewer dealers hold the bond, customers seeking to buy face fewer inventory-holding competitors. Ask-side competition can weaken. At the same time, more dealers have room to buy from customer-sellers, so bid-side competition can strengthen. The bid and ask prices may both rise, but for opposite reasons. Their difference—the bid-ask spread—can therefore change very little even though the economics of intermediation have changed substantially.
Dealers must also balance the profits earned on the two sides. Carrying inventory creates opportunities to sell to customer-buyers, but it uses capacity that could have been reserved for buying from customer-sellers. The model summarizes the choice through four linked components: ask-side profits, bid-side profits, the liquidity yield earned from holding the asset, and the dealer's net cost or benefit of carrying it. When one component changes, the others adjust until dealers are again willing to participate on both sides.
Order flow is part of this balance. A spread is a per-trade margin; profit also depends on how many trades arrive. In the paper's calibration, order flow is about 14 trades per year per dealer per side. That means a comparatively small change in a one-way spread can accompany a much larger change in side-specific profits if the volume of opportunities or the distribution of inventory changes. Reading only the average quote can hide the incentive that keeps a dealer in the market.
This is why the title's claim is narrower than “corporate-bond liquidity did not improve.” Electronic competition can genuinely reduce the cost of executing a trade. The warning is about inference. A narrower spread does not, by itself, show that investors face no holding-period compensation, that dealer capacity is abundant, or that the market will absorb a wave of selling under stress.
The evidence also has a defined scope. The paper is a preliminary working paper and its numerical results come from a calibration focused on investment-grade bonds with trades larger than $1 million between January 2017 and March 2021, excluding the crisis month of March 2020. Its normal-times model fixes the number of dealers and abstracts from dealer exit, one-sided crisis order flow, risk, asymmetric information, and adverse selection. Those omissions do not weaken the mechanism inside its intended setting; they mark the questions that a narrow spread cannot answer on its own.
To see exactly how the two liquidity measures separate, the next step is to examine what each one captures and how a customer's future trading need enters the bond's value.