Extracted main text — title through conclusion, appendix excluded. This is what our citation measures are computed over, published so the extraction can be checked by eye.
80,347 characters · 26 sections · 44 citation commands
On-chain Peak Shaving
In electrical grid management, peak shaving describes the practice of shifting consumption away from high-demand periods to reduce marginal cost exposure. Grid operators have spent decades developing the institutions, pricing signals, and scheduling tools that make this possible: real-time pricing markets, demand response contracts, and interruptible load programs. Firms that master peak shaving gain a durable cost advantage over those that pay spot rates indiscriminately.
Blockchain networks present an economically analogous problem. Gas fees — the computational charges levied on every Ethereum transaction — follow a congestion pricing logic nearly identical to real-time electricity pricing. When network demand surges, a dynamic base-fee mechanism raises the cost of block inclusion for all users simultaneously, regardless of whether their individual transactions contributed to the congestion. The surge is often driven not by operational activity but by speculative trading: arbitrage bots, NFT launches, and decentralized exchange activity that fills blocks above their target capacity and imposes a cost externality on every concurrent user. A pharmaceutical firm recording a drug shipment, a remittance platform settling a payment corridor, and a tokenized real-estate platform closing a property transfer all pay higher fees in the same block as a wave of speculative trades — not because their operations are more complex, but because they share infrastructure with actors whose demand is both larger and more elastic.
Do firms recognize this structure and respond to it? Using transaction-level on-chain data from seven firms from seven industries — spanning technology, fintech, healthcare, supply chain, real estate, consumer goods and administrative services — over a twelve-month window from May 2025 to April 2026, we find that they do. Firms in our sample systematically cluster their Ethereum transactions in off-peak hours, avoiding the European and U.S. business-hours windows when gas fees are highest. Peak costs occur reliably between hours 11 and 18 UTC (6 AM to 1 PM ET), with the single most expensive window at hour 15 (10 AM ET), where the hour-of-day premium reaches 0.220 USD above the overnight baseline — statistically significant at the 99.9% confidence level. The firms in our sample are, in the language of energy operations, peak shaving.
Yet peak shaving on-chain is harder than its grid management analog, for a structural reason that has no direct parallel in electricity markets: on-chain order flow is sparse. A manufacturing firm or healthcare provider does not generate continuous, schedulable transaction volume the way an industrial facility draws continuous power. Transactions are episodic, triggered by shipments, settlements, compliance deadlines, and counterparty actions that cannot always be deferred. As a result, even firms that demonstrably avoid peak windows still find themselves transacting during high-cost periods more often than a pure cost-minimization strategy would predict. Peak shaving mitigates on-chain cost exposure; it does not eliminate it. The residual cost burden — the gap between what disciplined scheduling achieves and what full avoidance would achieve — is a permanent operational friction that existing blockchain adoption frameworks have not characterized.
This paper makes four contributions. First, we document and name on-chain peak shaving as a firm behavior: the systematic scheduling of blockchain transactions toward low-congestion windows to reduce gas fee exposure. To our knowledge, this is the first empirical demonstration that operational firms exhibit this behavior across multiple industries on a public blockchain. Second, we extend Transaction Cost Economics (TCE) to account for time-varying execution costs imposed by network congestion externalities — a cost category absent from Williamson's original taxonomy and from existing blockchain-in-operations studies. Third, we introduce a typological distinction between transactional and speculative-arbitrage on-chain demand that explains the mechanism generating intraday cost heterogeneity: speculative actors fill blocks during business hours, raising base fees for operational users who share the network. Fourth, we develop an On-Chain Scheduling Matrix that translates our empirical findings into actionable guidance for operations managers evaluating transaction timing strategies across different urgency and cost-sensitivity profiles.
The remainder of the paper is structured as follows. Section 2 extends TCE to incorporate intraday congestion costs and positions our contribution against the closest antecedents in the operations management literature. Section 3 develops the measurement framework, including the transaction fullness measure, the transactional/speculative decomposition, and the intraday cost model. Section 4 describes the data and sample construction. Sections 5 and 6 present the findings, first at the pooled network level and then firm by firm across industries. Section 7 discusses theoretical and managerial implications. Section 8 concludes.
Transaction Cost Economics (TCE) provides the theoretical backbone of this paper. Originating with coase1937nature and given its canonical formulation in williamson1985economic, the framework organizes the costs of economic exchange into three phases: ex-ante costs (search, information-gathering, contracting), execution costs incurred at the moment of exchange, and ex-post costs (monitoring, enforcement, dispute resolution). The appropriate governance structure is determined by transaction attributes, most importantly asset specificity, uncertainty, and frequency. Asset specificity is the primary driver since high specificity creates bilateral dependency that favors hierarchical or contractual governance over spot markets williamson1985economic.
Critically, Williamson treats execution costs as stable conditional on governance structure: wire-transfer fees and clearing charges do not fluctuate by an order of magnitude within a single business day, so the framework assigns them no strategic role. Williamson also singles out maladaptation costs, the costs arising when a governance arrangement cannot adjust to changing circumstances. The maladaptation costs are similarly treated as bilateral, emerging from the contractual relationship between transacting parties and are addressable through renegotiation. Both assumptions fail in public blockchain environments, requiring the targeted extension we develop in Section (ref).
TCE has been applied productively to digital exchange since malone1987electronic predicted that electronic markets would shift governance toward market-mediated exchange by reducing coordination costs. gurbaxani1991impact formalized this logic, showing that IT investments alter firm boundaries by differentially reducing internal versus external coordination costs. Subsequent empirical work confirmed that asset specificity and uncertainty jointly predict governance choices, while also documenting a complementarity between formal contracts and relational governance that the original framework did not anticipate poppo2002substitutes. These contributions establish that digital technologies transform the composition of transaction costs rather than eliminate them, a finding our blockchain evidence extends to a new class of execution-phase costs.
The operations management literature on blockchain has developed rapidly since babich2020distributed provided the field's first systematic map of blockchain's operational strengths and weaknesses. Early work concentrated on traceability and transparency francisco2018supply, tian2017supply, sylim2018blockchain, logistics applications hackius2017blockchain, kouhizadeh2021blockchain, and financial services guo2016blockchain, peters2016understanding. A recurring finding is that implementation costs have been systematically underestimated: onboarding costs across supply chain tiers are routinely overlooked, and consortium formation and smart-contract governance costs can equal or exceed the savings from disintermediation in the early adoption phase babich2020distributed. Our paper extends these findings to the execution-cost dimension: even after successful deployment, time-varying gas fees create an ongoing cost burden that existing adoption frameworks have not characterized.
More recent theoretical contributions have engaged directly with governance costs. lumineau2021blockchain established that blockchain constitutes a third mode of organizing collaboration that is distinct from both contractual and relational governance. This collaboration works by mechanically enforcing agreements without requiring court enforcement or repeated-interaction trust. halaburda2024digitization extend TCE specifically to smart contracts as a governance mode, arguing that strong smart contracts trade adaptability for verifiability. Both contributions identify governance substitution effects but treat cost reduction as relatively uniform across settings. Our cross-industry evidence shows that this uniformity assumption fails: cost outcomes depend heavily on when firms transact, not just on whether they have adopted blockchain.
Three recent empirical contributions further discipline our claims: klockner2022blockchain provide event-study evidence that blockchain announcements generate positive abnormal returns concentrated in supply chain and logistics applications; cui2023value demonstrate that optimal blockchain design varies substantially across supply chain structures; and wang2019blockchain apply affordance theory to explain the systematic mismatch between blockchain adoption expectations and implementation reality. Our finding that firms engage in active on-chain scheduling is direct evidence of affordance recognition: these firms have learned that gas fee timing is a lever they can pull.
The concept of peak shaving originates in electrical grid economics, where it describes the practice of shifting consumption away from high-demand periods to reduce exposure to marginal cost spikes under real-time pricing strbac2008demand, schweppe1988spot. The structural parallels between electricity congestion pricing and blockchain gas fees are not merely rhetorical:
The technical economics of Ethereum gas fees have received growing attention outside OM journals. The EIP-1559 fee market mechanism Buterin2021EIP1559 introduced a dynamic base fee adjusted algorithmically to target 50% block utilization, partially reducing volatility while preserving priority-fee competition for block position; Roughgarden2021 establishes that EIP-1559 is incentive-compatible for myopic miners and improves fee predictability relative to the prior first-price auction. daian2020flash formalized the Maximal Extractable Value (MEV) problem as the ability of block producers to reorder or insert transactions to extract value from other users. Our transactional--speculative decomposition builds on this directly: MEV extraction is a primary driver of the speculative demand that fills blocks during business hours and imposes costs on operational users. john2024ethereum show that infrastructure costs in decentralized finance create structural inefficiencies that constitute permanent on-chain cost floors, supporting our argument that peak-shaving costs are a permanent operational problem rather than an adoption-phase artifact.
shang2023need provide the closest methodological antecedent to our study. Using a queuing-theoretic model applied to Bitcoin transaction data, they show that transaction fee dynamics are structured rather than random, a premise our paper builds on directly. Where shang2023need study a single platform and focus on the fee-speed tradeoff from a network design perspective, we study operational firms across seven industries on Ethereum and focus on the scheduling response from a firm cost-management perspective.
Table (ref) positions this paper relative to the streams reviewed above. Three gaps motivate this study. First, no prior work has documented on-chain peak shaving as a firm behavior or characterized intraday gas fee dynamics from the perspective of operational users rather than network designers; existing fee literature treats firms as passive price-takers, while our data show they are active schedulers. Second, while individual industry applications have received substantial attention, no comparative cross-industry analysis within an OM lens systematically maps how industry characteristics moderate on-chain cost exposure and scheduling flexibility; the uniformity assumption implicit in existing blockchain-in-OM studies is an empirical claim our seven-industry dataset tests and rejects. Third, TCE has not been extended to account for time-varying execution costs imposed by congestion externalities: prior extensions of TCE to digital exchange address how IT changes the level of transaction costs across governance structures, but not how IT creates within-period cost variation that itself has strategic implications. This paper addresses all three gaps in a unified empirical framework.
Transaction Cost Economics distinguishes three cost phases: ex-ante costs incurred before exchange (search, contracting, auditing), execution costs incurred at the moment of exchange, and ex-post costs incurred after exchange (monitoring, enforcement, and dispute resolution). Williamson's original formulation treated execution costs as relatively stable, conditional on the governance structure. At the time of williamson1975markets, wire-transfer fees, clearing charges, and settlement costs did not fluctuate by an order of magnitude within a single business day. On public blockchains, they do. We therefore extend TCE along a single, targeted dimension: gas fees are execution costs whose magnitude is endogenously determined by aggregate network demand at the moment of transaction submission, introducing a time-varying component absent from the original framework.
Formally, let $C_{it}$ denote the total execution cost incurred by firm $i$ in submitting a transaction at time $t$. Under the EIP-1559 fee mechanism Buterin2021EIP1559, this decomposes as:
where $g_{it}$ is gas consumed, $B_t$ is the protocol-determined base fee at block $t$, $\pi_{it}$ is the priority fee offered by the firm, and $M_{it}$ is the firm's declared maximum fee per gas unit. The base fee $B_t$ is not under the firm's control: it is determined by the fullness of the preceding block and adjusts upward when aggregate demand exceeds the protocol target.
This structure has a direct TCE interpretation. The component $g_{it}\cdot B_t$ is a congestion externality cost imposed on firm $i$ by unrelated network participants. In Williamson's taxonomy, costs imposed by unrelated third parties with no contractual relationship to the transacting firm are most naturally classified as maladaptation costs: the firm cannot renegotiate the base fee after block inclusion, cannot contractually insulate itself from congestion, and cannot recover overpayments if network conditions ease after submission. The component $g_{it}\cdot\pi_{it}$ is a discretionary execution cost that the firm does control, and whose optimization is the subject of the scheduling response we study empirically. We therefore introduce a dual classification: gas fees are execution costs in timing but maladaptation costs in origin. This is a distinction absent from the original framework and from existing blockchain-in-operations studies babich2020distributed, lumineau2021blockchain.
\paragraph{Boundary condition.} This extension applies whenever three conditions hold simultaneously: (i) the network employs dynamic marginal-cost pricing, so $B_t$ is endogenous to aggregate demand; (ii) the firm's transaction volume is small relative to total network demand, so the firm is a price-taker with respect to $B_t$; and (iii) at least a subset of the firm's transactions is deferrable within an operationally acceptable window, so a scheduling response is feasible. When any condition fails—on permissioned chains with administratively fixed throughput costs, for a dominant network participant capable of influencing $B_t$, or for a firm whose entire transaction population is time--critical—the standard TCE execution cost treatment remains appropriate.
We define on-chain peak shaving as the systematic scheduling of blockchain transactions toward low-congestion time windows to reduce gas fee exposure. The behavior is rational whenever: (i) the base fee $B_t$ exhibits predictable intraday variation; (ii) at least some transactions are deferrable within the intraday window without material operational cost; and (iii) the cost savings from deferral exceed the operational cost of delay.
Condition (i) follows from the EIP-1559 mechanism, which ties $B_t$ to block fullness $\phi_b = G^{\mathrm{used}}_b / G^{\mathrm{max}}_b$, where $G^{\mathrm{used}}_b$ is gas consumed and $G^{\mathrm{max}}_b$ is the block gas limit. When $\phi_b > 0.5$, the base fee rises by up to 12.5% per block; when $\phi_b < 0.5$, it falls proportionally Roughgarden2021. Because transaction demand inherits the intraday periodicity of global financial markets, European and U.S. business hours generate predictably higher base fees than overnight or weekend windows.
For condition (ii), we introduce a binary deferrability indicator $d_{it} \in \{0,1\}$, where $d_{it}=1$ if transaction $t$ can be submitted in any block within a window $[t, t+\Delta]$ without breaching a contractual, regulatory, or operational deadline, and $d_{it}=0$ otherwise. Plain value transfers and routine record-keeping are typically deferrable ($d_{it}=1$); compliance-driven filings, DeFi liquidations, and counterparty-triggered closings are typically not ($d_{it}=0$).
Condition (iii) requires that $g_{it}\cdot\Delta B > \kappa_i$, where $\Delta B = B_{\mathrm{peak}} - B_{\mathrm{off\text{-}peak}}$ is the base-fee differential and $\kappa_i$ is the firm's operational cost of delaying a deferrable transaction by one window. For high-gas transactions (complex contract interactions, multi-hop settlements, batch oracle updates), the left-hand side is large, making peak shaving worthwhile even when delay costs are non-trivial. For plain value transfers ($g_{it} \approx 21{,}000$ gas), the threshold is reached only at sufficient transaction volume or when $\Delta B$ is large.
These conditions yield four testable propositions:
A central identification challenge is that block fullness $\phi_b$ (and therefore the base fee firms seek to avoid) aggregates two qualitatively distinct types of demand. A transaction is transactional if it constitutes a necessary operational step in the delivery of a good, service, or financial instrument; transactional demand is characterized by low price elasticity, predictable periodicity tied to business cycles, and addresses attributable to identifiable commercial entities. A transaction is speculative arbitrage if its primary purpose is to exploit price discrepancies across venues or information states; this demand is characterized by high price elasticity, burst clustering around price events, and addresses attributable to algorithmic trading bots, DEX routers, or MEV searchers daian2020flash.
Block fullness is partitioned as follows:
where $\phi^T_b$, $\phi^S_b$, and $\phi^U_b$ denote the shares of block gas attributable to transactional, speculative arbitrage, and unclassified activity, respectively. Each tagged address is assigned a primary activity type $\delta_a \in \{T, S\}$ based on: (i) the contract's verified source code and ABI; (ii) the distribution of counterparty addresses; and (iii) the priority-fee distribution of outgoing transactions relative to the block-level 75th-percentile tip, with addresses consistently submitting above-threshold tips classified as $\delta_a = S$. Including $\phi^S_b$ as a control variable in subsequent regressions isolates the cost pass-through borne by transactional firms from the congestion that they did not generate.
\paragraph{Intraday cost structure (Propositions 1--2).} The primary estimating equation regresses realized gas fees on hour-of-day indicators in a cross-sectional panel:
where $\mathit{Gas}_{it}$ is the realized gas cost in USD; $\mathbf{1}[H_{it}=h]$ is an indicator for the UTC hour of submission, with hour 23 as the omitted baseline; $\hat{\phi}^S_{b(t)}$ is the speculative demand share of the containing block; $\mu_i$ are firm fixed effects; and $\lambda_w$ are week fixed effects. Standard errors are HC3-robust throughout. The coefficients $\hat{\beta}_h$ identify the hour-of-day cost premium relative to the overnight reference window; under Proposition 1 we expect $\hat{\beta}_h > 0$ for $h \in [11, 18]$ UTC and $\hat{\beta}_h \approx 0$ or negative for $h \in [0, 6]$ UTC.
\paragraph{Firm-level avoidance ratios (Proposition 2).} To test whether firms exhibit active peak-shaving behavior rather than incidental off-peak activity, we construct an avoidance ratio for each firm $i$:
where low-cost hours are defined as those with $\hat{\beta}_h < 0$ in the pooled regression, and equivalently as the Peak Shaving Score:
$\mathit{PSS}_i > 0$ indicates that firm $i$ concentrates transactions in off-peak hours beyond the uniform-scheduling benchmark of $16/24 \approx 66.7\%$. Statistical significance is assessed using a permutation test: for each firm, transactions are randomly reassigned across the 24-hour clock in 10,000 replications, and observed $A_i$ values outside the 95th percentile of this null distribution constitute evidence of active peak shaving.
\paragraph{Residual cost burden (Proposition 4).} Even firms with $A_i > 1$ incur a residual cost burden relative to a counterfactual of perfect off-peak scheduling:
where $\bar{C}_i$ is firm $i$'s observed mean gas cost per transaction and $\bar{C}^*_i = N_i \cdot \overline{\mathit{Gas}}_{i,h^*}$ is the counterfactual cost if all transactions had been submitted during the cheapest observed hour $h^*_i = \arg\min_h \overline{\mathit{Gas}}_{i,h}$. $R_i \geq 0$ by construction; a large $R_i$ relative to $\bar{C}_i$ indicates that peak shaving is constrained by transaction sparseness.
\paragraph{Cross-industry comparison (Propositions 3--4).} Because the sample contains seven firms, formal regression of $A_i$ on TCE moderators is underpowered. We instead use structured qualitative comparative analysis MHS19, scoring each firm on transaction frequency, asset specificity, and deferrability using a three-point rubric developed a priori from the TCE literature and mapping the resulting profiles to observed $A_i$ and $R_i$ values. This approach is consistent with established practice in theory-building OM research with small cross-industry samples Yin18, Eis89.
Another readily available proxy for congestion can be the total block reward collected by the block miner for each block. The more transactions in the block, the higher the congestion. Also, a higher number of transactions in the block results in a higher total reward that the miner gains from the block.
Figure (ref) shows the hourly statistics for the total block rewards. As the figure shows, once again, the U.S. business hours see the highest total block rewards, with the dispersion of the rewards spiking around the U.S. markets daily close. These results are consistent with the institutional trading patterns, suggesting that many institutions are at the forefront of the technology.
Transaction-level records were extracted from Etherscan.io via API for the period January 1 through March 31, 2026. Each record contains: transaction hash; block number; Unix timestamp (UTC); originating and destination addresses; smart-contract address (where applicable); transaction fee in ETH and USD; the prevailing USD-to-ETH exchange rate; and an execution-status indicator. Block-level total validator reward $R_b$ was downloaded from Etherscan.io separately for each block appearing in the transaction data.
The analysis covers seven firms drawn from seven industries (Table (ref)). Firms were selected on the basis of sustained, identifiable on-chain activity during the sample window; several candidates (Open Campus, I\c{s}iklar Holdings) had no recorded activity in 2026 and were excluded. The final sample contains $N = 62{,}142$ confirmed transactions after removing failed transactions (isError $= 1$) and records missing any required field.
\paragraph{Block-level congestion proxy.} We measure the block fullness by the relative block reward $R_b$ that the validator receives from assembling the block. Because $R^{\max}_b$ adjusts dynamically toward a protocol target of $G^* = 15 \times 10^6$ gas units, we additionally construct a block-reward congestion proxy:
where the 95th-percentile ceiling $\hat{R}^{(95)}$ prevents extreme MEV-extraction blocks from compressing the scale for ordinary operational blocks.
\paragraph{Transaction-type classification.} Each record is classified into one of three mutually exclusive types based on the input and methodId fields: eth (plain value transfer, input $= \texttt{0x}$, fixed base cost of exactly 21,000 gas); call (external contract call, variable gas); or deploy (contract creation, \texttt{to} $= \emptyset$). Records in our sample carry a single \texttt{gasPrice} field, indicating legacy (Type-0) transactions. Under this model, the total fee is $F_t = g_t \cdot p_t$ with $p_t$ expressed in Wei.
Firm-level summary statistics are reported in Table (ref).
Table (ref) shows the hour-of-the-day dummy variable regressions. As the table shows, the hours of 11-18 (6 AM ET - 1 PM ET) have consistently higher transaction costs, reflecting the network congestion during European and U.S. business hours. The hours of 10 and 11 AM ET see the highest transaction costs. The Adjusted $R^2$ of 1.6% demonstrates weak but positive explanatory power of hour-of-the-day dummy variables. Measuring Gas in USD uncovers similar time-of-day patterns, yet lower Adjusted $R^2$.
Figure (ref) shows the intraday patterns for the average and standard deviation of gas fees. The figure also shows the proportion of trades in our sample occurring at different hours. Figure (ref) shows that the highest average costs occur during the European and U.S. trading hours, peaking at 15 (10 AM ET). The standard deviation of costs, however, rises most at h=20 (3 PM ET, near the close of the U.S. markets). Figure (ref) documents that, when pooled, transactions in our sample occur during the Asian session, which has the lowest fee structure. The next section examines firm-by-firm responses to the transaction costs.
Coins.ph is a Philippines-based fintech platform that routes remittance flows between overseas Filipino workers in the United States, the Middle East, Singapore, and Hong Kong and their recipients in the Philippines, substituting on-chain settlement for the correspondent banking chain. With $N = 54{,}651$ transactions over the sample window, it is the largest firm in the sample by transaction count and the closest empirical approximation to the textbook definition of peak shaving: high-frequency, low-gas-intensity, and largely deferrable plain ETH value transfers. Figure (ref) and Table (ref) summarize the intraday transaction details for Coins.ph.
Anchorage Digital is a federally chartered digital asset bank providing institutional custody, trading, and settlement services. Its 1,785-transaction sample reflects the settlement leg of institutional digital asset transactions, including custody transfers, internal allocations, and inter-institutional settlements. These transactions are subject to T+1 regulatory timelines on certain instruments that constrain transaction deferrability, making it a canonical low-deferrability firm in the On-Chain Scheduling Matrix (Table (ref)). Table (ref) and Figure (ref) summarize the results.
Propy Inc.\ is a blockchain-based real estate platform that uses Ethereum smart contracts to automate property closing workflows, including escrow initialization, title verification, payment settlement, and deed recording. Each property closing generates a discrete, episodic cluster of high-gas contract interactions rather than continuous order flow, and closing deadlines, counterparty coordination requirements, and regulatory recording windows constrain the firm's ability to shift transactions to off-peak hours even when the intraday cost schedule would make deferral economically attractive. Propy's transactions in our sample are summarized in Figure (ref) and Table (ref).
The Nike (Ondo) entry represents a tokenized equity instrument rather than an operational consumer goods deployment: Ondo Finance has issued a tokenized version of Nike stock as a 1:1 digital wrapper around underlying shares held with regulated custodians. With only 72 transactions over the sample window, this is the smallest firm by transaction count; its inclusion illustrates how tokenized financial instruments generate an on-chain cost profile driven by market events rather than operational scheduling. These transactions are summarized in Figure (ref) and Table (ref).
BrainTrust is a decentralized talent network that executes job matching, payment enforcement, credential verification, fee distribution, and governance management through Ethereum smart contracts, making every routine administrative action an on-chain transaction subject to prevailing gas fees. Its smart contracts are by design devoid of token value transfers, which is why the Value field is structurally missing from many of its Etherscan records, and its transaction timing is governed by governance and payroll cycles anchored to U.S.\ business hours without operational necessity. Braintrust's transactions are summarized in Figure (ref) and Table (ref).
Solve.Care is a healthcare coordination platform that uses the Ethereum blockchain to improve administrative efficiency, data interoperability, and care-coordination workflows, with on-chain activity consisting primarily of care-coordination confirmations, administrative sign-offs, and routine record-keeping. With 116 transactions over the sample window it is the second-smallest firm by count, but its intraday scheduling pattern provides the clearest visual illustration of active peak-shaving behavior in the sample: transaction activity peaks at $h = 6$ UTC while the highest gas costs are recorded during U.S.\ business hours. Solve.Care intraday transaction evolution is shown in Table (ref) and Figure (ref).
Morpheus.Network is a Canadian SaaS middleware platform that connects legacy enterprise systems like ERP systems, customs agencies, logistics providers, and IoT sensors into a blockchain-backed workflow engine. Morpheus.Network native token validates and executes supply-chain workflow steps on-chain. Its IoT-oracle contract calls and multi-hop settlement interactions are computationally heavier than plain value transfers, placing it in the high-gas-intensity segment of the sample and amplifying the dollar cost impact of block congestion relative to low-gas-intensity peers. Morpheus.Network intraday transaction dynamics are shown in Table (ref) and Figure (ref).
The empirical findings in Section (ref) establish two firm-level dimensions that jointly determine how much cost reduction is available from transaction timing. The first is transaction deferrability $d_{it} \in \{0,1\}$: whether a given transaction can be submitted in any block within a window $[t, t+\Delta]$ without breaching a contractual, regulatory, or operational deadline. The second is gas intensity $g_{it}$: the computational weight of the transaction, which determines how large the dollar savings from avoiding a peak-hour base-fee premium actually are. Together, these two dimensions generate four scheduling regimes formalized in the On-Chain Scheduling Matrix (Table (ref)).
The regime boundaries are empirically grounded. The threshold separating statistically significant from insignificant hour-of-day premia is taken directly from Table (ref): hours 3, 5, and 6 UTC have $|\hat{\beta}_h| < 0.015$ with $|t| < 0.7$, making them the lowest-cost submission windows in the sample. The peak-hour premium of \$0.220 at hour 15 UTC (10\,AM ET) is significant at the 99.9% confidence level and represents the upper bound of the cost schedule a firm faces. Operations managers can therefore treat the regression coefficients in Table (ref) as a forward price curve for gas, updated as new data accumulate, and use the inequality $g_{it} \cdot \Delta B > \kappa_i$ to dynamically determine whether a given transaction belongs in Regime I or Regime II as its gas estimate and the prevailing fee schedule become known.
Where peak shaving shifts costs in time, Regime III introduces cost provisioning as a scheduling substitute for firms with low deferrability. A firm that has internalized the hour-of-day premium schedule in Table (ref) can build a gas budget that accounts for the expected \$0.220 surcharge on transactions submitted between hours 11--18 UTC, rather than treating gas fees as an unforecastable variable cost. This reframing from uncontrollable externality to forecastable input cost is the central managerial implication of the extended TCE framework developed in Section (ref), and positions gas-fee management alongside energy procurement and foreign exchange hedging as a domain where operations managers can apply systematic forward-planning discipline even when real-time scheduling is infeasible.
Table (ref) summarizes Peak Shaving Scores, fee savings, residual cost floors, and fullness pass-through coefficients for all seven firms. Three findings stand out.
\paragraph{Scheduling discipline is externally constrained, not uniformly absent.} Positive PSS values are concentrated among firms whose geographic footprint or transaction type creates a natural off-peak scheduling opportunity: Solve.Care ($\text{PSS} = +0.014$), Coins.ph ($+0.006$), and Nike/Ondo ($0.000$, effectively neutral). Negative scores characterize firms whose timing is governed by external parties: settlement counterparties (Anchorage Digital, $-0.097$), property closing deadlines (Propy, $-0.048$), or governance cycles tied to U.S.\ business hours (BrainTrust, $-0.076$; Morpheus.Network, $-0.028$). These firms need to transact regardless of whether gas intensity is high or low. Solve.Care's positive score is notable because healthcare carries high asset specificity, which Proposition 3 predicted would reduce avoidance behavior; the finding suggests that operational geography moderates the asset specificity effect in ways that the original TCE framework does not capture.
\paragraph{Residual cost floors are structural, not scheduling failures.} Even firms with positive PSS values cannot eliminate peak-hour cost exposure. Residual floors range from 40.7% of actual expenditure (Propy) to 92.5% (Nike/Ondo), and even the cheapest scheduling hour for each firm carries non-zero block-reward fullness ($\widehat{\varphi}^{br}_{i,h^*}$ ranging from 0.184 to 0.462; Table (ref)). This confirms Proposition 4: the execution cost floor is a structural feature of the Ethereum network. Speculative-arbitrage demand maintains a non-zero base level of block fullness even during the lowest-congestion hours, and the externality it imposes on operational users cannot be eliminated by scheduling discipline alone. The two largest absolute floors, Coins.ph (\$4,209) and Propy (\$513), reflect different mechanisms: high transaction volume in the former, and high gas intensity and low deferrability in the latter. These artifacts illustrate that the floor is determined by whichever constraint is binding.
\paragraph{Pass-through heterogeneity maps onto asset specificity.} The fullness pass-through coefficient $\widehat{\delta}_i$ varies by more than an order of magnitude across firms, from $-0.245$ (Nike/Ondo) to $0.578$ (Propy), and maps cleanly onto the TCE dimension of asset specificity. Propy's real estate contract interactions are highly asset-specific: each transaction encodes unique property identifiers, counterparty addresses, and legal instrument references. Each transaction then carries the highest pass-through accordingly. Anchorage Digital's standardized custody transfers yield a low pass-through ($0.069$). BrainTrust's administrative calls are the least asset-specific and produce a pass-through indistinguishable from zero ($0.024$, $p = 0.242$). Nike/Ondo's negative coefficient reflects its speculative-arbitrage demand profile: tokenized equity transactions cluster around price dislocations rather than in response to congestion, reversing the expected positive relationship between block fullness and gas costs.
\paragraph{Weekday--weekend differentials reinforce this picture.} As Table (ref) documents, Anchorage Digital and Propy both show significant weekday premiums (22.4% and 24.4%, respectively, $p < 0.001$), while Coins.ph's intraday scheduling discipline substantially neutralizes the day-of-week distinction at the fee level, despite a significant underlying fullness differential $\Delta\widehat{\varphi}^{br} = 0.071$, $p < 0.01$.
This paper documents and theorizes on-chain peak shaving: the systematic scheduling of blockchain transactions toward low-congestion time windows to reduce gas fee exposure. We measure on-chain peak shaving as a firm behavior observable across multiple industries on a public blockchain. Drawing on transaction-level data from seven firms across seven industries on the Ethereum network from January through March 2026 ($N = 62{,}142$ transactions), we establish four findings.
Gas fees exhibit statistically significant and economically meaningful intraday variation: the hour-15 UTC premium of \$0.220 per transaction is significant at the 99.9% confidence level and is driven primarily by speculative-arbitrage demand rather than operational activity. Firms respond actively: Peak Shaving Scores are positive for firms with geographic or operational scheduling flexibility and negative for firms whose transaction timing is governed by external deadlines or institutional settlement cycles. This confirms that firms have learned to treat the intraday fee schedule as a manageable cost input rather than an unforecastable externality. This scheduling capacity is heterogeneous and moderated by two firm-level dimensions: transaction deferrability and gas intensity. These dimensions jointly determine both the recommended timing strategy and the magnitude of achievable savings, as formalized in the On-Chain Scheduling Matrix. Even disciplined schedulers cannot eliminate peak-hour cost exposure entirely: residual cost floors ranging from 40.7% (Propy) to 92.5% (Nike/Ondo) of actual gas expenditure establish that a permanent execution cost floor exists, reflecting both the sparseness of operational transaction flow and the non-zero block congestion that persists even during the cheapest scheduling hours.
Theoretically, these findings require an extension of Transaction Cost Economics to account for time-varying execution costs imposed by congestion externalities. The dual classification of gas fees as execution costs in timing but maladaptation costs in origin is the central theoretical contribution: it identifies a cost category absent from Williamson's original taxonomy and from existing blockchain-in-operations studies, and it positions on-chain gas-fee management alongside energy procurement and foreign exchange hedging as a domain requiring systematic operational planning. Practically, the evidence shows that the operations management community has underestimated the strategic importance of transaction timing in blockchain deployments. Existing adoption frameworks evaluate blockchain primarily on whether it reduces ex-ante and ex-post transaction costs; our evidence shows that the execution cost component varies by more than 100% within a single business day and is partially controllable through scheduling discipline. Blockchain is not a universal cost-reduction technology but a governance substitution mechanism whose net cost benefit must be evaluated empirically, case by case, accounting for industry-specific cost composition and firm-specific scheduling capacity.
Four directions merit priority in future research. Longitudinal studies tracking on-chain cost evolution as Layer-2 adoption expands would establish whether the intraday peak structure documented here persists as network architecture matures. Primary survey data from blockchain operations practitioners would ground the deferrability classifications inferred here in direct managerial accounts. Environmental sustainability costs, such as energy consumption and carbon intensity per transaction, are a natural extension of the TCE framework developed here and would enable a fuller accounting of the social costs of blockchain adoption SBFK20. Finally, behavioral dimensions of smart-contract adoption, including the role of cognitive load and bounded rationality in managers' scheduling decisions, represent a rich agenda for behavioral operations management research at the frontier of blockchain adoption.