1. Uniswap V2 and the Bonding Curve

In Uniswap V2, each pool holds reserves of two tokens, EURC and USDC, denoted by $ x $ and $ y $ , which satisfy the constant-product invariant

(1) $$ k=x\times y. $$

The marginal price of EURC in terms of USDC is $ {P}_{x/y}=y/x $ . Because reserves and transactions are public on-chain, price setting is deterministic and liquidity is continuously available as long as both tokens remain in the pool.

Swaps move reserves along the bonding curve defined by (1). If a trader purchases $ \Delta x>0 $ EURC in exchange for USDC, post-trade reserves satisfy

(2) $$ {x}^{\prime }=x-\Delta x,\hskip1em {y}^{\prime }=\frac{k}{x^{\prime }},\hskip1em {P}_{x/y}^{\prime }=\frac{k}{{\left(x-\Delta x\right)}^2}. $$

Since $ {x}^{\prime }<x $ , the marginal price rises mechanically, generating slippage that increases with trade size. Liquidity mints instead scale up $ k $ and shift the bonding curve outward. To keep $ {P}_{x/y} $ unchanged, LPs add both tokens in proportion to the prevailing price so that $ y/x $ remains constant. Burns reverse this process.

Figure 2 illustrates the relationship between swaps, liquidity provision, and price impact. Graph (a) provides a conceptual illustration in which swaps move the pool along a given bonding curve from $ {E}_0 $ to $ {E}_1 $ , while liquidity mints shift the curve outward by increasing available reserves at the prevailing price, moving the pool from $ {E}_0 $ to $ {E}_2 $ . Graph (b) then proceeds with a numerical example showing how price impact varies with pool depth for a given trade size.Footnote ⁹

FIGURE 2EURC/USDC Bonding Curves and Slippage

Graph A of Figure 2 illustrates bonding curves and liquidity provision in Uniswap. Starting from an initial equilibrium $ {E}_0 $ , a swap trade that purchases EURC moves the pool along the bonding curve to $ {E}_1 $ , while an increase in liquidity at the prevailing price shifts the equilibrium from $ {E}_0 $ to $ {E}_2 $ . Graph B illustrates the relationship between pool size and price impact in a constant-product automated market maker for the EURC/USDC pair. The left subgraph shows bonding curves for different liquidity levels, while the right subgraph shows percentage price impact for trades of varying size (in EURC). Price impact is lower in more liquid pools, reflecting reduced slippage.

Each pool in Graph (b) starts at $ {P}_{x/y}=1.10 $ but differs in scale. For a low-liquidity pool with $ \left(x,y\right)=\left(\mathrm{100,110}\right) $ , a purchase of $ \Delta x=5 $ raises the price to $ {P}_{x/y}^{\prime}\approx 1.219 $ , while the same trade yields $ {P}_{x/y}^{\prime}\approx 1.122 $ and $ 1.111 $ in medium- and high-liquidity pools, respectively.

More generally, the percentage price impact of a trade of size $ \Delta x $ is

(3) $$ \mathrm{Price}\ \mathrm{Impact}\left(\%\right)=100\times \left[\frac{1}{{\left(1-\frac{\Delta x}{x}\right)}^2}-1\right], $$

which depends on trade size relative to pool depth, with larger pools exhibiting flatter curves and smaller price responses.

2. Uniswap V3 and Concentrated Liquidity

Uniswap V3 generalizes the constant-product design by allowing LPs to concentrate liquidity within a user-defined price range $ \left[{p}_a,{p}_b\right] $ rather than distributing it uniformly over $ \left[0,\infty \right) $ . This improves capital efficiency by ensuring that liquidity is active only when the market price lies within the specified range. V3 also introduces multiple fee tiers (0.01%, 0.05%, 0.3%, and 1%) that segment liquidity across pools and allow LPs to choose different risk–return profiles.Footnote ¹⁰

Within any active range, trades are priced using virtual reserves that satisfyFootnote ¹¹

(4) $$ \left(x+\frac{L}{\sqrt{p_b}}\right)\left(y+L\sqrt{p_a}\right)={L}^2, $$

where $ L $ is the liquidity parameter governing effective depth within the active range. A higher $ L $ implies deeper liquidity and smaller price impact. When the range expands to $ \left[0,\infty \right) $ , (4) reduces to $ x\times y={L}^2 $ , so $ L $ coincides with the Uniswap V2 constant-product parameter.

3. Tick-Based Pricing and Liquidity Distribution

V3 discretizes prices into ticks indexed by $ i\in \mathrm{\mathbb{Z}} $ , where $ {p}_i={1.0001}^i $ . Tick spacing determines which ticks can host liquidity. In the EURC/USDC 0.05% pool, tick spacing is 10, so liquidity is allocated in 10-tick intervals. Figure 3 plots tick-level liquidity for this pool at block 19,771,559 (Apr. 30, 2024). Liquidity is concentrated near the market price and tapers toward the tails, resembling a tokenized limit order book. Further details on liquidity aggregation and price setting are provided in Sections B.1 and B.2 of the Supplementary Material.

FIGURE 3Snapshot of EURC/USDC Liquidity

Figure 3 shows the tick-level distribution of liquidity around the prevailing pool price for the EURC/USDC 0.05% Uniswap V3 pool. The horizontal axis reports tick distance from the current market price (tick 0), where each tick corresponds to a discrete price interval in log base $ \sqrt{1.0001} $ . The pool has a fixed tick spacing of 10, implying price intervals of approximately 10 basis points. Ticks to the left of zero represent liquidity below the current price and correspond to buy limit orders for EURC, while ticks to the right represent liquidity above the current price and correspond to sell limit orders for EURC. Liquidity at each tick is expressed in 1000 of EURC.

1. CLS EUR/USD Benchmark and Uniswap EURC/USDC Price

We source a benchmark EUR/USD rate from CLS, which provides a volume-weighted average price of interbank quotes at 5-min intervals. We aggregate these data to hourly and daily frequency. EURC/USDC prices are constructed as the last transaction price at each UTC hour (and day) from the Uniswap V3 EURC/USDC pool, obtained via the Subgraph API.Footnote ¹²

Graph A of Figure 4 compares EURC/USDC and EUR/USD prices and plots their difference. Consistent with Adams et al. (Reference Adams, Lader, Liao, Puth and Wan2023), the DEX rate tracks the benchmark closely, with an average absolute deviation of 24 basis points. Deviations are larger early in the sample when liquidity is low, so our empirical analysis begins on Aug. 15, 2022. A major episode is the March 2023 USDC de-pegging following concerns about reserve exposure to SVB; during Mar. 11–12, 2023, EURC/USDC traded at a relative premium versus EUR/USD.

FIGURE 4EURC/USDC Prices

Figure 4 plots EURC/USDC and EUR/USD prices at the hourly frequency. EURC/USDC prices are sourced from Uniswap V3, and EUR/USD prices are sourced from CLS. Graph A shows the EURC/USDC price, the benchmark EUR/USD price, and their cross-market price difference. Graph B reports cumulative blockchain order flow and prices in the EURC/USDC market. The left subgraph shows aggregate order flow, prices, and liquidity provision in EURC and USDC, while the right subgraph disaggregates cumulative order flow by trader group, including sophisticated traders (Top 10 wallets), primary dealers (PM), liquidity providers (LPs), their intersections, and all remaining traders. The sample period for Graph A is from June 28, 2022 to Apr. 30, 2024, and for Graph B is from Aug. 15, 2022 to Apr. 30, 2024.

2. DEX Trading Volume and Liquidity Provision

Our Uniswap V3 data set contains the complete history of swap transactions in the EURC/USDC market. Each swap is recorded at the wallet level, where a wallet corresponds to an Ethereum address controlling the associated tokens.Footnote ¹³ We complement swap data with proprietary Kaiko data on liquidity provision, which record mint and burn transactions by LPs, including the quantities of EURC and USDC added or removed and the price ranges over which liquidity is allocated.Footnote ¹⁴

Using wallet histories, we classify market participants into three base groups reflecting trading activity and access to issuance and liquidity provision. Sophisticated traders are defined as the top 10 wallet addresses by trading volume in each month, consistent with the microstructure view that large trades are more likely to reflect informed or institutional participation (Barber, Odean, and Zhu (Reference Barber, Odean and Zhu2008), Easley and O’Hara (Reference Easley and O’Hara1987)). These addresses account for an average of 52% of total trading volume between Aug. 15, 2022 and Apr. 30, 2024. Primary dealers are wallets that transact directly with the EURC or USDC treasury, minting stablecoins by depositing fiat and redeeming them at par.Footnote ¹⁵ LPs are wallets that supply or withdraw liquidity by minting and burning positions in the EURC/USDC pool. Each of the latter two groups accounts for approximately 7% of aggregate trading volume.

Table 2 reports summary statistics for seven trader groups. These consist of the three base groups, their pairwise intersections, and a residual category. The base groups contain 76 sophisticated traders, 68 primary dealers, and 90 LP addresses. Six addresses are classified as Top 10 $ \cap $ primary dealers, seven as Top 10 $ \cap $ LPs, and three as primary dealer $ \cap $ LPs.Footnote ¹⁶ The remaining 2342 addresses form the residual category. Additional details on wallet characteristics and trading and liquidity provision behavior are provided in Section C of the Supplementary Material.

TABLE 2Trader Classification

Blockchain order flow. We construct signed order flow as net buyer-initiated EURC volume. Each swap reports amount0 (EURC) and amount1 (USDC) from the pool’s perspective. Since EURC is the base currency, amount0 < 0 indicates a buyer-initiated EURC trade, while amount0 > 0 indicates a sale of EURC to the pool. For interval $ t $ ,

(5) $$ {OF}_t=\sum \limits_{k\in \mathcal{K}(t)}\left(\unicode{x1D7D9}\;\left[{T}_k=B\right]-\unicode{x1D7D9}\left[{T}_k=S\right]\right)\times {V}_k, $$

where $ \mathcal{K}(t) $ indexes swaps in $ t $ , $ {T}_k $ indicates buyer or seller initiation for EURC, and $ {V}_k $ denotes EURC-equivalent trade size.

Graph B of Figure 4 plots cumulative order flow alongside prices, showing positive comovement. Disaggregating by trader group reveals that Top 10 traders often take the opposite side of LPs and residual wallets. This pattern is consistent with an asymmetric-information interpretation in which sophisticated traders exploit short-lived mispricing, while LPs and residual wallets supply liquidity on the other side, a channel we test formally in Section IV.

Turning to LPs, their net EURC purchases during periods of EUR appreciation reflect inventory rebalancing rather than informed trading. As swap traders buy EURC from the pool, LP inventories become increasingly tilted toward USDC. LPs respond by withdrawing liquidity from stale price ranges, acquiring EURC through swaps, and re-minting liquidity at updated price bounds. This behavior is documented at the transaction level in Section C.3.3 of the Supplementary Material and is consistent with LPs passively absorbing order flow and rebalancing inventories, in a manner analogous to inventory management by traditional FX dealers (Bjønnes and Rime (Reference Bjønnes and Rime2005), Lyons (Reference Lyons1995)).

Liquidity measurement. To characterize the supply-side response to trading activity, we measure net liquidity using Uniswap V3 mint and burn events following Klein, Kozhan, Viswanath-Natraj, and Wang (Reference Klein, Kozhan, Viswanath-Natraj and Wang2024). These events record additions and withdrawals of liquidity over price intervals $ \left[{P}_a,{P}_b\right] $ at the block level. For each block $ k $ ,

(6) $$ {\displaystyle \begin{array}{c}{mint}_{(k)}^{net}={mint}_{(k)}^{ask}-{mint}_{(k)}^{bid},\\ {}{burn}_{(k)}^{net}={burn}_{(k)}^{ask}-{burn}_{(k)}^{bid},\\ {}{Liquidity}_{(k)}^{net}={mint}_{(k)}^{net}-{burn}_{(k)}^{net}.\end{array}} $$

A positive $ {Liquidity}_{(k)}^{net} $ indicates that more ask-side (EURC-side) liquidity is added than withdrawn in block $ k $ . To distinguish between active and passive liquidity provision, we classify positions within $ \pm 1\% $ of the mid-price as best liquidity and positions outside this range as away. We aggregate block-level liquidity measures to the hourly frequency to align them with order flow and returns. Section B.3 of the Supplementary Material provides full construction details.

3. Additional Data

CLS volume. To study traditional-market activity, we use the CLS FX Spot Flow data set, which settles around 40% of global FX volume across spot, swaps, and forwards and covers 18 currencies.Footnote ¹⁷ The data provide hourly buy and sell volumes between banks and three categories of price takers (funds, nonbank financials, corporates) and are widely used in FX microstructure research (Huang et al. (Reference Huang, O’Neill, Ranaldo and Yu2023), Kloks, Mattille, and Ranaldo (Reference Kloks, Mattille and Ranaldo2023), and Ranaldo and Somogyi (Reference Ranaldo and Somogyi2021)). We construct sector-level volumes for interbank, bank–funds, bank–nonbank financials, and bank–corporates, with interbank volume obtained by subtracting bilateral bank–client flows from the aggregate series.

Figure 5 compares hourly Uniswap EURC/USDC activity with EUR/USD CLS volumes by sector. Traditional trading is concentrated between 13:00 and 16:00 UTC and is driven mainly by interbank and fund–bank flows, coinciding with the overlap of London, Frankfurt, and New York trading hours and the 16:00 UTC WMR fix (Krohn, Mueller, and Whelan (Reference Krohn, Mueller and Whelan2024)). Blockchain trading occurs throughout the day, with peaks in the afternoon and around 09:00 UTC, consistent with 24/7 trading and retail participation. Average daily CLS volume is 28.42 billion EUR, compared to 0.423 million EURC on Uniswap, or about 0.0015% (0.15 bps) of total EUR/USD trading. Additional summary statistics and intraday liquidity patterns are reported in Section C.3.2 of the Supplementary Material and Table 3.

FIGURE 5Hourly FX Trading Volume

Figure 5 plots trading volume at the hourly frequency. Graph A reports trading volume on Uniswap V3 for the EURC/USDC market, expressed in 1000 of EURC. Graph B reports trading volume on CLS for the EUR/USD market, disaggregated by counterparty sector (bank–bank, bank–fund, bank–corporate, and nonbank financial–bank) and expressed in EUR millions. The sample period is from Aug. 15, 2022 to Apr. 30, 2024.

TABLE 3Summary Statistics: Prices, Volume, and Blockchain Variables

Transaction costs and volatility. Gas fees compensate Ethereum validators for confirming transactions. We retrieve transaction-level gas fees for each swap from Etherscan (in ETH, converted to USDC) and use them to compute arbitrage bounds and trader-level transaction costs.Footnote ¹⁸ To capture market-wide conditions, we use the EthVol index from T3 Index as a measure of expected 30-day implied volatility for Ether and a daily gas-fee index from CoinMetrics to study price efficiency at the daily frequency.Footnote ¹⁹

Intermediary constraints. We use two measures of intermediary constraints. The first is the intermediary capital risk factor (ICRF) from He, Kelly, and Manela (Reference He, Kelly and Manela2017), defined as AR(1) innovations to the market-based capital ratio of U.S. primary dealers. The second follows Huang et al. (Reference Huang, Ranaldo, Schrimpf and Somogyi2025) and captures violations of the law of one price (VLOOP) in G10 FX markets, constructed from minute-level LSEG quotes on EUR/USD, USD/X, and EUR/X cross rates.Footnote ²⁰ The first principal component of standardized VLOOP series explains about 46% of total variation and captures global inter-dealer pricing distortions. Summary statistics are reported in Table 3.

Fact #1: DEX Prices and Volumes Are Closely Connected to Traditional FX Markets

We begin by documenting how decentralized markets comove with traditional FX benchmarks in both prices and quantities. Our baseline measure of price alignment is the absolute deviation between EURC/USDC and the CLS EUR/USD benchmark,

(7) $$ {\Delta}_t=\left|{p}_{\mathrm{EUR}\mathrm{C}/\mathrm{USD}\mathrm{C},t}-{p}_{\mathrm{EUR}/\mathrm{USD},t}\right|. $$

The average deviation is 24 basis points, with a median of 16 basis points, and exceeds 200 basis points only during the March 2023 USDC de-pegging episode. These moments indicate that decentralized prices generally track the benchmark closely, although short-lived dislocations arise under stress.

Trading volumes on decentralized and traditional venues also display a strong connection. DEX activity follows the intraday pattern of CLS volumes, with a clear peak between 13:00 and 16:00 UTC when European and U.S. markets overlap. Using CLS volumes disaggregated into interbank, nonbank financial, and corporate sectors, and comparing them to DEX activity across sophisticated traders, primary dealers, and LPs, we find that decentralized trading mirrors the pattern of the interbank segment, which the literature identifies as the most informed part of the FX market (Huang et al. (Reference Huang, O’Neill, Ranaldo and Yu2023), Ranaldo and Somogyi (Reference Ranaldo and Somogyi2021)). Section D.1 of the Supplementary Material reports the full set of volume correlations across participant groups and trading hours, confirming that correlations are strongest with the interbank segment and during the core trading window when New York, London, and Frankfurt are open.

Formally, we estimate the relationship between decentralized and traditional FX volumes at the hourly frequency,

(8) $$ {V}_{N_{DEX},t}=\alpha +\sum \limits_{i\in {N}_{CLS}}{\beta}_i{V}_{N_{CLS},t}+{\varepsilon}_t, $$

where $ {V}_{N_{DEX},t} $ denotes DEX trading volume by participant type, measured in EURC, and $ {V}_{N_{CLS},t} $ denotes CLS sector-level trading volume, measured in EUR millions. Table 4 shows that interbank trading is most strongly associated with DEX activity. For sophisticated traders, a €1 million increase in interbank CLS volume is associated with approximately 4.35 units of EURC trading on DEXs.

TABLE 4DEX and CLS Volume Correlations

Figure 6 provides additional evidence. Graph A shows that DEX volumes for all trader categories peak between 13:00 and 16:00 UTC and decline outside these hours, with volumes falling by about 50% for sophisticated traders and 37% for primary dealers. For traders classified as both sophisticated and primary dealers, the drop reaches 74%. Graph B shows that weekend activity falls significantly across groups, with the steepest decline of 87% among sophisticated primary dealers. Together, these findings demonstrate that decentralized FX markets move closely with traditional FX markets across both price and quantity dimensions.

FIGURE 6Weekend and Weekday Volume by Trader Type

Figure 6 plots average trading volume, distinguishing between weekday and weekend trading by trader group. Graph A compares weekday trading volume during traditional primary-market opening hours (13:00–16:00 UTC) with other weekday hours. Graph B compares average trading volume across weekdays and weekends. Volumes are expressed in EURC. Blockchain trading volume is disaggregated by trader group following Table 2. The sample period is from Aug. 15, 2022 to Apr. 30, 2024.

Fact #2: Peg Efficiency Is Driven by Blockchain Frictions

In an efficient market, decentralized prices should track benchmark EUR/USD values tightly. Persistent deviations therefore reveal the frictions that limit arbitrage. To identify these frictions, we regress $ {\Delta}_t $ on blockchain execution costs, stablecoin fundamentals, and traditional intermediary constraints at the daily frequency,

(9) $$ {\displaystyle \begin{array}{c}{\Delta}_t={\beta}_0+{\beta}_1{\mathrm{gasfee}}_t+{\beta}_2{\sigma}_{ETH,t}^{IV}+{\beta}_3{R}_{ETH,t}+{\beta}_4\mid {p}_{\mathrm{USDC}/\mathrm{USD},t}-1\mid \\ {}+{\beta}_5\mid {p}_{\mathrm{EURC}/\mathrm{EUR},t}-1\mid +{\beta}_6{\mathrm{VLOOP}}_t+{\beta}_7{\mathrm{ICRF}}_t+{\varepsilon}_t,\end{array}} $$

where $ {\mathrm{gasfee}}_t $ , $ {\sigma}_{ETH,t}^{IV} $ , and $ {R}_{ETH,t} $ capture on-chain execution costs and crypto-market risk, $ \mid {p}_{\mathrm{USDC}/\mathrm{USD},t}-1\mid $ and $ \mid {p}_{\mathrm{EURC}/\mathrm{EUR},t}-1\mid $ measure stablecoin peg deviations, VLOOP is a standardized measure of FX triangular arbitrage violations, and ICRF is the intermediary capital risk factor of He et al. (Reference He, Kelly and Manela2017), which proxies shocks to U.S. dealer balance-sheet constraints.

Table 5 shows that blockchain-specific frictions are the primary determinants of price deviations. Both gas fees and Ether volatility consistently explain variation in $ {\Delta}_t $ . A $1 increase in gas fees raises $ {\Delta}_t $ by about 1.3 basis points, while a 1-basis-point increase in implied Ether volatility increases deviations by 0.13–0.16 basis points. Ether volatility matters not because of dealer balance-sheet constraints but because it raises uncertainty for traders holding cryptoasset portfolios, reducing their willingness to deploy capital for arbitrage.

TABLE 5Determinants of EURC-USDC Peg Deviations

Stablecoin fundamentals also play an important role. The significance of $ \mid {p}_{\mathrm{USDC}/\mathrm{USD}}-1\mid $ reflects spillovers from USDC market conditions into EURC/USDC pricing. Frictions in USDC/USD markets constrain arbitrage capital for participants benchmarking EURC/USDC to EUR/USD, and as the more liquid leg of the pair, USDC/USD conditions provide a more informative proxy for these constraints. This pattern is partly explained by the SVB de-pegging event of March 2023, during which Circle’s disclosure of $3.3 billion in SVB deposits caused USDC to temporarily trade below par, generating sharp spikes in $ {\Delta}_t $ .Footnote ²¹

Consistent with this mechanism, column 5 shows that a 1-basis-point increase in the USDC/USD peg deviation is associated with a 0.75-basis-point rise in $ {\Delta}_t $ , whereas deviations in the EURC/EUR peg are positive but statistically insignificant. To assess whether this effect holds outside the de-pegging episode, column 6 re-estimates the full specification excluding the 4 days of the SVB episode (Mar. 10–13, 2023). The $ {R}^2 $ falls from 0.23 to 0.08 and the coefficient rises to 1.84 with a larger standard error, reflecting more limited variation in the USDC peg outside the event, but the coefficient remains statistically significant.

All other coefficients remain stable across columns 5 and 6, confirming that the core blockchain-friction results are not driven by the SVB episode. Traditional intermediary constraints such as VLOOP and ICRF are insignificant, suggesting that balance-sheet frictions binding intermediaries in OTC FX markets do not carry over to decentralized venues. Taken together, these results indicate that price inefficiencies in the EURC/USDC market arise primarily from blockchain-specific frictions (execution costs, on-chain congestion, and stablecoin fundamentals) rather than traditional intermediary constraints (Barbon and Ranaldo (Reference Barbon and Ranaldo2024), Foley et al. (Reference Foley, O’Neill and Putniņš2023)).

Fact #3: Transaction Costs Vary by Trader Type; Gas Fees Dominate for Most, Slippage for Large Traders

We next examine transaction costs at the trade level. Total cost, expressed in basis points relative to trade size, is the sum of gas fees, LP fees, slippage, and private-routing fees. Gas fees are payments to validators for processing transactions on the public blockchain and are largely fixed per transaction, implying higher proportional costs for small trades. LP fees are constant at 5 basis points in the EURC/USDC pool. Slippage reflects the price impact of consuming liquidity along the bonding curve and is measured as the difference between the execution price and the pool price immediately before the trade. Private-routing fees arise only for privately routed transactions and represent payments to validators for authenticating these orders; they are approximated using transfers to validator addresses, as detailed in Section C.2.4 of the Supplementary Material. Across trader types, median total costs range between 20 and 50 basis points.

Figure 7 shows substantial heterogeneity in trading costs across participants. Graph A indicates that larger traders face lower and more concentrated cost distributions, while smaller or residual wallets incur higher and more dispersed costs. Graph B decomposes median total costs by component. Gas fees are the dominant cost for most users, accounting for roughly 7–23 basis points across groups and representing a larger fraction of total costs for smaller traders. Slippage becomes more important for larger trades, reaching up to about 20 basis points, while LP fees are constant by design at 5 basis points. Overall, median total trading costs range from around 15–17 basis points for LP- and PM-related addresses to roughly 45 basis points for Top 10 $ \cap $ LP wallets, with Top 10 traders facing median costs just above 30 basis points.

FIGURE 7Trading Costs

Figure 7 presents trading cost measures for the EURC/USDC market at the hourly frequency. Graph A reports the interquartile range of total transaction costs across trader groups. Total costs combine gas fees (Ethereum transaction fees converted to USD), private fees (payments to validators for privately routed transactions), liquidity provider fees (5 basis points for the EURC/USDC pool), and price slippage, all expressed in basis points. Graph B decomposes median transaction costs into these components by trader group. Trader classifications follow Table 2. The sample period is from Aug. 15, 2022 to Apr. 30, 2024.

These results have two main implications. First, trading costs define the effective bounds within which arbitrage can operate. Section D of the Supplementary Material shows that once gas fees, slippage, LP fees, and private-routing fees are incorporated, the share of arbitrage-bound violations is approximately 16%–19%, falling to 3%–5% after accounting for centralized exchange fees and OTC bid–ask spreads. The EURC/USDC market is therefore best characterized as constrained price efficient rather than fully frictionless.

Second, decentralized trading costs compare favorably with those faced by many OTC clients. While on-chain costs exceed inter-dealer spreads (EUR/USD spreads averaged only 0.55 basis points in 2023 (Filippou, Maurer, Pezzo, and Taylor (Reference Filippou, Maurer, Pezzo and Taylor2024)), they are of a similar order of magnitude to those paid by less-privileged OTC clients. Hau et al. (Reference Hau, Hoffmann, Langfield and Timmer2021) show that FX derivatives clients at the 90th percentile face spreads of up to 50 basis points. Liquidity depth remains limited, particularly for large trades, but decentralized venues may nonetheless expand access to FX trading for smaller participants.

Public Information

We next examine trading responses to public information. These episodes provide clean settings to study the arbitrage mechanism, as common shocks generate price dislocations across centralized and on-chain venues. We focus on two types of public information: the March 2023 USDC de-pegging episode and scheduled Federal Reserve announcements.

USDC de-pegging event. The USDC de-pegging on Mar. 11, 2023 provides a natural setting to study arbitrage under stress. When it became public that $3.3 billion of USDC reserves were held at SVB, which had entered resolution, concerns about reserve accessibility caused USDC to trade at a discount and fall to 87 cents on centralized exchanges. Confidence recovered on March 13, after the Federal Deposit Insurance Corporation announced that all SVB deposits would be fully protected.Footnote ²²

Figure 8 plots deviations of EURC/USDC from the EUR/USD benchmark together with disaggregated blockchain order flow. Sophisticated traders are net buyers of EURC throughout the episode, while primary dealers and smaller wallets exhibit net selling pressure. This behavior is consistent with arbitrage opportunities created by the USDC de-pegging. As USDC traded at a discount on centralized exchanges, sophisticated traders could purchase USDC on those venues and sell it into Uniswap V3 pools where its relative price was higher, thereby arbitraging the temporary dislocation. LPs behaved differently. Their swap and mint–burn activity did not respond to price discrepancies, and their inventories adjusted mechanically as traders executed swaps, consistent with passive liquidity provision. Section E.1 of the Supplementary Material provides additional evidence of repeated cross-market arbitrage across multiple pools, shows that cost and routing advantages favored Top 10 wallets, and documents minimal liquidity re-positioning during the episode.

FIGURE 8USDC De-Pegging Event: Blockchain Order Flow of Different Trading Groups

Figure 8 plots cumulative blockchain order flow around the USDC de-pegging event at the hourly frequency. Price deviations are measured as the difference between EURC/USDC prices from Uniswap V3 and EUR/USD prices from CLS. Order flow ( $ OF $ ) measures net buyer transactions of EURC and is sourced from Uniswap V3 trade data. Cumulative order flow is disaggregated by trader group following Table 2. The sample period is from Mar 10, 2023 to Mar 12, 2023.

Monetary announcements. Public information also arrives through scheduled macroeconomic releases. Section E.2 of the Supplementary Material examines Federal Reserve FOMC announcements and shows that EURC/USDC deviations from CLS benchmarks remain small after releases, while trading volumes rise sharply in both markets. Prices in the two venues move together almost immediately, indicating active arbitrage as macroeconomic news is incorporated.

Taken together, the regression evidence and event studies show that arbitrage linking decentralized and traditional FX prices is concentrated among sophisticated traders and primary dealers, consistent with scale economies in blockchain execution costs, while other traders respond weakly to short-horizon price gaps and LPs remain largely passive.

Permanent Price Impact

We study whether blockchain order flow contains information about exchange rate fundamentals beyond arbitrage by examining its permanent price impact. In asymmetric-information models, public news is incorporated rapidly, whereas private information is revealed gradually through order flow (Evans and Lyons (Reference Evans and Lyons2002)). In traditional FX markets, this process operates primarily through inter-dealer trading. Blockchain markets lack such an inter-dealer tier, so informational trading is transmitted directly through trades rather than through dealer inventories. In this sense, informed trading on chain more closely resembles the framework of Kyle (Reference Kyle1985), in which traders with private signals reveal information through their trades.

The presence of informed trading is likely to differ across participants. Primary dealers, with direct access to EUR and USD funding and links to interbank markets, are plausibly more exposed to fundamental signals. Sophisticated traders, operating at scale and facing lower execution costs, can transmit such information across venues. By contrast, LPs primarily supply inventory and are exposed to adverse selection when prices move on fundamentals (Capponi and Jia (Reference Capponi and Jia2025)). This heterogeneity motivates an empirical design that estimates permanent price impact separately by trader type.

We estimate a structural VAR at the hourly frequency to capture the dynamic relationship between order flow and exchange rate changes. To isolate the informational content of blockchain order flow beyond what is already incorporated through institutional trading, we control for traditional FX order flow from CLS. This yields a block SVAR with traditional OTC order flow ( $ {\mathbf{OF}}_t^{OTC} $ ), blockchain order flow ( $ {\mathbf{OF}}_t^{DEX} $ ), and exchange rate changes ( $ \Delta {p}_t $ ):

(11) $$ \left[\begin{array}{c}{\mathbf{OF}}_t^{OTC}\\ {}{\mathbf{OF}}_t^{DEX}\\ {}\Delta {p}_t\end{array}\right]=\alpha +\sum \limits_{k=1}^L{\mathbf{A}}_k\left[\begin{array}{c}{\mathbf{OF}}_{t-k}^{OTC}\\ {}{\mathbf{OF}}_{t-k}^{DEX}\\ {}\Delta {p}_{t-k}\end{array}\right]+{\epsilon}_t. $$

The vector $ {\mathbf{OF}}_t^{OTC} $ includes buy-minus-sell imbalances for interbank dealers, funds, nonbank financials, and corporates from CLS data. The vector $ {\mathbf{OF}}_t^{DEX} $ consists of EURC/USDC transaction flows on Uniswap, disaggregated by wallet type: LPs, residual wallets ( $ \notin \left\{\mathrm{Top}10,\mathrm{PM},\mathrm{LP}\right\} $ ), Top10 $ \cap $ LP, PM, Top10, and Top10 $ \cap $ PM, as defined in Section II. The variable $ \Delta {p}_t $ denotes log changes in the EUR/USD exchange rate, measured either from the DEX mid-price or the CLS benchmark.

Identification uses a recursive Cholesky ordering $ {\mathbf{OF}}_t^{OTC}\to {\mathbf{OF}}_t^{DEX}\to \Delta {p}_t $ . This allows traditional OTC order flow to affect blockchain order flow and prices contemporaneously, while DEX flows do not contemporaneously affect OTC flows, and prices respond immediately to all order flows. The ordering reflects the hierarchical structure of FX trading venues and standard information transmission assumptions in the microstructure literature. Full details are provided in Section F of the Supplementary Material.

Figure 9 reports impulse response functions over a 24-hour horizon. The figure shows responses of EURC/USDC returns and benchmark EUR/USD returns from CLS to shocks in DEX order flow by participant type. Order flow from sophisticated traders and primary dealers generates the largest and most persistent effects on benchmark returns, whereas flows from LPs have negligible impact.

FIGURE 9Price Impact of Blockchain Order Flow by Trader Groups

Figure 9 plots impulse responses of returns to a 1 million EURC shock in blockchain order flow, estimated at the hourly frequency. The figure reports responses for transactions by Top 10 wallets, primary dealers, other wallets, and liquidity providers, for both EURC/USDC returns from Uniswap V3, and benchmark EUR/USD returns from CLS. Responses are estimated using a structural VAR with 1,000 bootstrap replications. The sample period is from Aug. 15, 2022 to Apr. 30, 2024.

Quantitatively, a 1 million EURC shock to primary dealer order flow produces a cumulative 24-hour impact of approximately 3.95% on CLS benchmark EUR/USD returns, while an equivalent shock to sophisticated trader order flow produces an impact of about 2.2%. The combined group of Top 10 wallets that are also primary dealers generates a cumulative impact of 3.1%. In contrast, order flow from LPs and residual wallets has negligible permanent effects on benchmark returns, with estimated impacts of approximately $ -0.05 $ and 0.2%, respectively, consistent with passive liquidity provision and uninformed trading.

Scaling these responses to a 1-standard-deviation shock in daily order flow implies permanent impacts of roughly 1.6 basis points for sophisticated traders and 2.9 basis points for primary dealers.Footnote ²³ These effects do not rely on inter-dealer inventory adjustment mechanisms. Instead, they are consistent with persistent price adjustment driven by heterogeneity in trader information and execution, in line with informed trading models such as Kyle (Reference Kyle1985).

Robustness. Section G of the Supplementary Material reports a full set of robustness checks. We re-estimate the baseline SVAR under alternative specifications, including controls for net liquidity provision (within and beyond $ \pm $ 1% of the best price), a higher-frequency (5-min) SVAR, and trader-volume quintiles. The main results are stable. Price impacts remain concentrated among top wallets and primary dealers, while LPs and smaller wallets exhibit weaker and less persistent effects. We also examine intraday variation (Section G.2 of the Supplementary Material) and find that price impacts are concentrated between 13:00 and 15:00 UTC, when major global financial markets are simultaneously open. Finally, we examine just-in-time liquidity provision (Capponi, Jia, and Zhu (Reference Capponi, Jia and Zhu2024)). Section G.3 of the Supplementary Material identifies one wallet, ending in “ae13,” that consistently displays this behavior in the EURC/USDC pool.Footnote ²⁴ This behavior is rare in our sample and does not materially affect our estimates of permanent price impact.