Pith. sign in

REVIEW 4 major objections 5 minor 47 references

Inferring Inventory Dynamics from Supply Chain Networks: A Graph Learning Approach with Autonomous Validation

T0 review · 4 major / 5 minor · reviewed 2026-07-14 · grok-4.5

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Supply-chain topology plus a production function can recover firm inventory changes, and multi-agent econometric checks can validate them without inventory labels.

desk verdict Solid production-constrained GNN for inventory changes; the multi-agent "verification without labels" claim is consistency theater, not accuracy. read the letter →

arxiv 2607.10642 v1 pith:RSUMXN7X submitted 2026-07-12 stat.AP

Tiancheng Gao , Xiang Zhang , Wei Lan , Bin Liu This is my paper
classification stat.AP
keywords Supply-DemandMismatchInventoryForecastingProductionFunctionGraphNeuralNetworksMulti-AgentReasoningSupplyChainSemi-SupervisedInference
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Many small and medium firms never publish inventory numbers, so supply-demand mismatch is hard to measure where it often matters most. This paper argues that inventory change is the cumulative gap between effective supply capacity and market demand, and that both can be recovered from who buys from whom in the supply network. A directed graph neural network, constrained by a simple production function that turns procurement into marketable output, predicts firm-level inventory changes from network structure and transaction flows, using listed firms as the only labeled anchors. Because SMEs still lack ground-truth labels for validation, five classical econometric models—spatial spillovers, dynamic persistence, causal direction, shock transmission, and supply-demand forecasting—are run as independent agents on the predicted series; an expert agent reconciles their findings into a single consistency judgment. The authors report stable forecast errors on held-out periods and show that the predicted dynamics line up with inventory-cycle logic and network transmission patterns across industries. If the approach holds, managers and policymakers can track mismatch and its propagation through industrial networks even when inventory books are missing.

What carries the argument

Production-function-constrained inventory update: effective supply Q = S(1+α) from procurement S, inventory change ΔI = Q − D + ε, with latent α and ε learned from directed receive/send embeddings; validated by five concurrent econometric agents whose conclusions an expert agent reconciles for economic consistency.

What would settle it

On a hold-out set of SMEs (or industries) that later disclose true inventory, check whether the graph model's predicted ΔI series are accurate by direct error metrics and whether the multi-agent consistency score actually rises and falls with that accuracy rather than merely producing coherent economic narratives.

Watch

Extended reading notes

Core claim

A production-function-constrained directed graph neural network can infer firm-level inventory changes from supply-chain topology and transaction flows, and a multi-agent econometric consistency framework can verify those predictions against established economic signatures even when ground-truth inventory observations for SMEs are unavailable.

Load-bearing premise

Agreement of the predicted inventory series with classical econometric patterns, as interpreted and reconciled by language-model agents, is treated as a valid substitute for ground-truth inventory labels when checking SME predictions.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes a production-function-constrained directed GNN that infers firm-level inventory changes from supply-chain topology, treating inventory as the gap between effective supply Q=S(1+α) and demand plus a latent adjustment (Eqs. 1–3, 16–18). Supervision uses Huber loss on listed firms with disclosed inventory (Eq. 21); SMEs are initialized from industry averages and inferred without labels. To address label scarcity, five classical econometric models (spatial panel, dynamic panel, cross-lagged, network DID, supply–demand forecasting) are wrapped as LLM agents that interpret industry aggregates of the predictions, with an expert agent reconciling inconsistencies. Empirical claims rest on stable RMSE/MAE/WAPE for 2021–2023 windows (Table 3), hyperparameter sensitivity (Figs. 3–4), cross-LLM agreement on industry driving type and inventory-cycle stage (Fig. 5), and qualitative industry-chain case studies (Figs. 6–8).

Significance. If the verification claim holds, the work would offer a practical route to firm-level inventory inference for SMEs where labels are systematically missing—an important and under-addressed problem in supply-chain analytics. Strengths include an explicit production-function and inventory-flow constraint inside a directed GNN (rather than a pure black-box topology model), a clear semi-supervised setup that uses listed firms as anchors, and a multi-perspective econometric consistency protocol with multi-LLM robustness checks. These design choices are creative and potentially useful for unlabeled network settings. The significance is currently limited by the gap between consistency-with-theory and accuracy, and by the absence of baselines and any external check on SME predictions.

major comments (4)
  1. Abstract, §5, and Conclusion claim the multi-agent framework enables “effective prediction verification even when ground-truth observations are unavailable.” §5.1–5.2 and Tables 1–2 implement this as agreement of industry-level predicted inventory with classical econometric signatures (spillovers, persistence, cross-lags, shock transmission, supply–demand forecasts), reconciled by LLM agents. Consistency with theory is not a substitute for accuracy: any series that preserves network co-movement and AR-like persistence can pass while remaining badly biased at the SME level. SME inventory is never observed and never held out; Eq. 24 aggregates to industry level and washes out firm-level error; cross-LLM agreement (Fig. 5) only shows similar interpretation of the same econometric outputs. The claim should be restated as economic-consistency screening, and the paper should either provide a l
  2. Table 3 reports RMSE/MAE/WAPE for 2021/06–2023/06 but does not state that evaluation is restricted to labeled (listed) nodes, nor does it include any baseline (industry-average inventory change, pure GNN without production constraint, AR/panel on listed firms, transaction-flow heuristics). Without baselines, “stable predictive performance” cannot establish that the production-function constraint or directed message passing is load-bearing. At minimum, add ablations removing α/ε structure (Eqs. 13–18) and directed receive/send embeddings (Eqs. 8–12), plus simple baselines, and clarify the evaluation population in Table 3.
  3. §4.1 and Eqs. 1–3, 16–18 define inventory change as ΔI = Q − D + ε with Q = S(1+α). The GNN is trained to recover inventory under that structure on listed firms, then industry aggregates are fed to econometric models that look for supply–demand, spillover, and persistence patterns (§5.1, Table 1). Part of the “alignment with economic theory” is therefore built into the generative structure rather than independently discovered. The paper should quantify how much of the multi-agent consensus is inherited from the inventory identity (e.g., by comparing agents on raw S−D series versus model-implied ΔI, or on predictions with α,ε frozen/random).
  4. §4.3–4.4 and Eq. 19 introduce several free scales (γ for SME and unlisted-at-t0 initialization, c_α for production adjustment, λ1 non-negativity, λ2 weight decay, Huber δ) that are only partially explored in Figs. 3–4. Text also conflicts on the chosen λ1 (0.2 in §6.1 vs 0.1 in the summary paragraph after Fig. 4). Because SME levels are never observed, sensitivity of industry-level agent conclusions to γ and to initial-inventory choices is load-bearing for the unlabeled claim and should be reported.
minor comments (5)
  1. §6.1: “λ1 = 0.2 and λ1 = 1×10−5” appears to be a typo for λ1 and λ2; later text uses λ1 = 0.1. Align all hyperparameter statements with the final configuration.
  2. Notation: both “SEMs” and “SMEs” appear (e.g., §2.2, Eq. 19); standardize to SMEs. “fist half of 2023” (case analysis) should be “first half.”
  3. Figure 1 and Figure 2 are referenced but their captions/legends in the text are thin; ensure receive/send paths and agent I/O are labeled so the architecture is reproducible without the full code.
  4. Table 3 sample sizes grow over time; clarify whether this is expanding firm coverage, listing events, or graph construction, and whether metrics are comparable across windows.
  5. Related work on inventory record inaccuracy and production networks is solid; a short comparison to other semi-supervised or physics-informed GNN inventory/operations papers would help position the contribution.

Circularity Check

3 steps flagged · score 4.0 of 10

Supervised inventory prediction is not circular, but multi-agent 'verification without labels' partly re-reads supply–demand and network structure already imposed by the inventory identity and GNN design.

  1. self definitional [§4.1 Eqs. (1)–(3), (16)–(18); §5.1 Table 1 (Dynamic Panel & Supply–Demand Forecasting)]
    "ΔIi(t)=Qi(t)−Di(t)+εi(t) … when inventory changes that cannot be directly observed are not considered, the inventory changes ΔIi(t) is equal to the difference between effective supply and demand. … Dynamic Panel Model Ykt=αYk,t−1+β1Skt+β2Dkt+… Evaluate inventory persistence (α) … Supply-Demand Forecasting Model Sk,t+1=γ1Ykt+λ1Skt+… Dk,t+1=γ2Ykt+λ2Dkt+…"

    Inventory change is defined as effective supply minus demand (plus adjustment), with Q=S(1+α). The Dynamic Panel and Supply–Demand Forecasting agents then treat co-movement of predicted Y with observed S and D, and the role of Y in forecasting S/D, as independent economic evidence. Any series built from that identity will mechanically exhibit the supply–demand linkages those agents are designed to detect, so 'consistency' is partly definitional rather than external validation of SME accuracy.

  2. self definitional [§4.2 node features; §5.1 Table 1 Dynamic Panel Model Agent]
    "The feature vector xi(t) comprises … (5) a momentum term capturing the lagged inventory change ΔIi(t−1). … 2. Dynamic Panel Model Ykt=αYk,t−1+β1Skt+β2Dkt+μk+εkt Evaluate inventory persistence (α) and assess whether industries exhibit sustained destocking or accumulation dynamics."

    Lagged inventory change is an explicit input to the GNN, so predicted dynamics inherit autoregressive structure by design. The Dynamic Panel agent then scores inventory persistence of the predicted (industry-aggregated) series as validation evidence—partially reading out structure that was fed in rather than independently confirming unlabeled forecasts.

1 more flagged steps
  1. self definitional [§4.3 directed message passing Eqs. (8)–(12); §5.1 Spatial Panel & Network DID agents]
    "In each propagation layer, a node first aggregates information from its upstream neighbors to update the receiving embedding and then aggregates information from its downstream neighbors to update the sending embedding. … Spatial Panel Model Ykt=ρ∑m≠k WkmtYmt+βXkt+… Identify spatial dependence in inventory changes … Network DID Model Ykt=βShockkt+γ∑m≠k WkmtShockmt+… Quantify both direct and network-mediated effects"

    Directed GNN aggregation forces firm-level predictions to depend on upstream/downstream neighbors; industry aggregates of those predictions therefore inherit network co-movement. Spatial Panel and Network DID agents then treat spatial/network dependence of predicted Y as evidence of economic plausibility. Detecting spillovers that message passing already induces is only weak, partly self-referential support for SME inventory accuracy.

full rationale

The core forecasting pipeline is ordinary supervised structural learning, not a closed derivation: listed-firm inventory is observed, the model is trained with Huber loss against those labels (Eq. 21), and Table 3 reports hold-out RMSE/MAE/WAPE on later periods. That part has independent content and is not circular. Circularity appears only in the stronger claim that the multi-agent econometric stack 'enables effective prediction verification even when ground-truth observations are unavailable.' Inventory is defined as ΔI = Q − D + ε with Q = S(1+α) (Eqs. 1–3, 16–18); lagged ΔI is an explicit node feature; and directed message passing couples neighbors. Industry aggregates of those outputs are then fed to agents whose stated objectives are precisely supply–demand co-movement, AR persistence, and network spillovers (Table 1; §5.1). Passing those checks is therefore partly encouraged by construction and is not an independent accuracy test for unlabeled SMEs. This is partial circularity in the validation claim (score 4), not a full reduction of the inventory forecasts themselves to their inputs (which would be 6–10). Weakness of consistency-as-accuracy is primarily a correctness risk, not a derivation collapse.

Assumptions & free parameters 8 free parameters · 6 assumptions · 3 invented entities

The load-bearing story rests on a simplified production/inventory identity, the claim that network embeddings identify latent efficiency and adjustment terms, arbitrary SME initialization scales, and—most critically—the methodological axiom that multi-agent econometric consistency (LLM-mediated) validates predictions without labels. Free parameters control initialization, loss trade-offs, and capacity of the GNN; invented latents α and ε are not independently measured.

free parameters (8)
  • γ SME initial-inventory scale
    SMEs initialized at 0.1× same-industry listed average inventory (§4.3 Eq. 19); chosen by hand, not estimated from SME data.
  • γ unlisted-at-t0 listed-firm scale
    Set to 0.5 when a listed firm was not listed at t=0 (§4.3); hand-chosen.
  • c_α production-adjustment scale
    Scales tanh output for α̂_i(t) (§4.3 Eq. 14); controls feasible efficiency range.
  • λ1 non-negativity penalty
    Weight on ReLU(−Î) in hybrid loss; reported as 0.2 via grid search and later as 0.1 in best-config summary (§6.1–6.2).
  • λ2 L2 weight decay
    Set to 1e-5 after sensitivity plots (§6.1–6.2).
  • Huber δ
    Robustness threshold fixed at 0.5 for supervised inventory loss (§4.4, §6.1).
  • GNN depth and embedding width
    Selected as 4 layers and 64-dim hidden size from sensitivity analysis (Fig. 3); architecture capacity is data-tuned.
  • Dropout rate
    Fixed at 0.2 on fusion and GNN layers (§6.1).
assumptions (6)
  • domain assumption Inventory change equals effective supply minus demand plus an unobserved adjustment: ΔI_i(t)=Q_i(t)−D_i(t)+ε_i(t).
    Imported from inventory-cycle / flow-conservation literature and used as the modeling identity (§3–4.1, Eqs. 2–3).
  • ad hoc to paper Effective supply is Q_i(t)=S_i(t)(1+α_i(t)) with α learned from network embeddings.
    Highly simplified production function; α absorbs labor, capital, losses, etc., without separate measurement (§4.1 Eq. 1, §4.3).
  • domain assumption Upstream/downstream network structure identifies latent α_i(t) and ε_i(t) for SMEs.
    Justified by citations on productivity spillovers and record inaccuracy, but not proven identifiable in this design (§4.3).
  • ad hoc to paper Cross-model econometric consistency of predicted inventory, as synthesized by LLM agents, validates predictions when labels are missing.
    Core unlabeled-validation premise of §5; equates theoretical plausibility with verification.
  • domain assumption Industry-level aggregation of firm ΔI and transactions preserves the supply-demand mismatch signal for cycle inference.
    Stated in §5.3 with exclusion of finance/real estate; used for all multi-agent outputs.
  • domain assumption Listed-firm inventory disclosures are accurate enough supervisory targets for Huber training.
    Implicit in L_sup (§4.4); paper elsewhere cites pervasive inventory record inaccuracy, creating tension.
invented entities (3)
  • Latent effective production rate α_i(t)
    purpose: Convert observed procurement into effective supply inside the GNN inventory recursion.
    Not measured from plant-level productivity data; inferred only from network embeddings and inventory fit on listed firms.
  • Latent inventory adjustment ε_i(t)
    purpose: Absorb shrinkage, record error, delays, and other unobserved inventory deviations.
    Defined as residual adjustment predicted from the same embeddings; no external measurement protocol.
  • Multi-agent econometric validation system (five model agents + expert review agent)
    purpose: Produce structured economic-consistency assessments of predicted inventory without ground truth.
    Methodological construct of the paper; outputs are LLM-mediated interpretations of standard econometric fits, not a new physical entity, but treated as an autonomous verifier.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Inferring Inventory Dynamics from Supply Chain Networks: A Graph Learning Approach with Autonomous Validation." pith.science (2026). https://pith.science/paper/RSUMXN7X

@misc{pith2026260710642,
  author       = {Pith},
  title        = {Pith review of: Inferring Inventory Dynamics from Supply Chain Networks: A Graph Learning Approach with Autonomous Validation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RSUMXN7X}},
  note         = {Machine review of arXiv:2607.10642}
}
read the original abstract

Supply-demand mismatch represents a fundamental challenge in supply chain management, yet its direct measurement remains particularly elusive for small and medium-sized enterprises (SMEs).These firms typically lack systematic inventory records, leaving labeled training data critically scarce. Conventional supervised learning methods rely heavily on labeled samples, rendering them ill-equipped to reliably validate firm-level predictions under such data-scarce conditions. To resolve this unlabeled-data dilemma, we develop a multi-agent semi-supervised inference framework that reframes the label-scarcity problem as a structured, collaborative task distributed across specialized agents. We first construct a production-function-constrained graph machine learning model that infers firm-level inventory changes directly from supply chain network topology. A dedicated econometric validation agent then concurrently loads five econometric models (spanning spatial spillovers, dynamic persistence, causal direction, shock transmission, and supply-demand forecasting) to generates structured economic evidence from complementary dimensions. An expert review agent synthesizes the structured econometric evidence and produces a unified consistency assessment by resolving cross-agent inconsistencies. Empirical results demonstrate stable predictive performance on inventory-change forecasting tasks. Multi-agent econometric validation further confirms that predicted inventory dynamics align closely with established economic theory in terms of causal structure and network transmission mechanisms. Critically, the proposed agent framework enables effective prediction verification even when ground-truth observations are unavailable.

Figures

Figures reproduced from arXiv: 2607.10642 by the authors.

Figure 1
Figure 1. Architecture of the Directed Graph Neural Networks for Inventory Changes Prediction [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. The Multi-Agent Econometric Framework for Validating Predicted Inventory Changes [PITH_FULL_IMAGE:figures/full_fig_p012_2.png] view at source ↗
Figure 3
Figure 3. Sensitivity Analysis of Embedding Dimension and Network Depth [PITH_FULL_IMAGE:figures/full_fig_p015_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Sensitivity Analysis of Hyperparameters λ1 and λ2 [PITH_FULL_IMAGE:figures/full_fig_p016_4.png]
Figure 5
Figure 5. Figure 5: Consistency Between Industry Driving Types and Inventory Cycles [PITH_FULL_IMAGE:figures/full_fig_p016_5.png]
Figure 6
Figure 6. Figure 6: Visualization of Inventory Changes in Coal & Consumable Fuels Industry Chain [PITH_FULL_IMAGE:figures/full_fig_p017_6.png]
Figure 7
Figure 7. Figure 7: Visualization of Inventory Changes in Renewable Electricity Industry Chain [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]
Figure 8
Figure 8. Figure 8: Visualization of Inventory Changes in Machinery & Supplies & Components Industry Chain [PITH_FULL_IMAGE:figures/full_fig_p018_8.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

47 extracted references · 17 canonical work pages

  1. [1]

    P. R. Winters. Forecasting sales by exponentially weighted moving averages.Management Science, 6(3):324–342, 1960

    1960

  2. [2]

    S. A. Roberts. A general class of holt-winters type forecasting models.Management Science, 28(7):808–820, 1982

    1982

  3. [3]

    E. S. Gardner and E. McKenzie. Forecasting trends in time series.Management Science, 31(10):1237–1246, 1985

    1985

  4. [4]

    A. Z. M. K. Harahap, M. K. I. A. Rahim, N. Malinjasari, S. M. Salleh, and R. A. Maarof. Enhancing inventory management through demand forecasting.International Journal of Research and Innovation in Social Science, 9 (1):2737–2744, 2025

    2025

  5. [5]

    Celestin, G

    M. Celestin, G. R. G. Raja, J. A. Mohamed, and D. M. Kumar. Mathematical models for demand forecasting in procurement: Balancing inventory and avoiding stockouts.International Journal of Advanced Trends in Engineering and Technology, 10(2):142–151, 2025

    2025

  6. [6]

    L. F. da Silva, E. S. Christo, K. A. Costa, and H. P. Tavares. Combining demand classification and forecasting mod- els in spare parts inventory management for the energy sector.International Journal of Advanced Manufacturing Technology, 2026

    2026

  7. [7]

    O. A. Olaniyi, P. S. Pugal, and M. Etim. Optimising inventory management strategies for cost reduction in supply chains: A systematic review.The Sankalpa: International Journal of Management Decisions, 12(1):97–103, 2026

    2026

  8. [8]

    Y . R. Sagaert and N. Kourentzes. Inventory management with leading indicator augmented hierarchical forecasts. Omega, 136:103335, 2025

    2025

Show all 47 references
  1. [9]

    Zizka and M

    M. Zizka and M. Kustka. Methodology for inventory management with sporadic demand in small and medium- sized enterprises.Acta Logistica, 13(1):174–185, 2026

    2026

  2. [10]

    Acemoglu, V

    D. Acemoglu, V . M. Carvalho, A. Ozdaglar, and A. Tahbaz-Salehi. The network origins of aggregate fluctuations. Econometrica, 80(5):1977–2016, 2012

    1977

  3. [11]

    V . M. Carvalho, M. Nirei, Y . U. Saito, and A. Tahbaz-Salehi. Supply chain disruptions: Evidence from the great east japan earthquake.The Quarterly Journal of Economics, 136(2):1255–1321, 2021

    2021

  4. [12]

    D. Wu, Q. Wang, and D. L. Olson. Industry classification based on supply chain network information using graph neural networks.Applied Soft Computing, 132, 2023

    2023

  5. [13]

    Huang, F

    S. Huang, F. Poursafaei, J. Danovitch, M. Fey, W. Hu, E. Rossi, J. Leskovec, M. Bronstein, G. Rabusseau, and R. Rabbany. Temporal graph benchmark for machine learning on temporal graphs. InProceedings of the 37th Annual Conference on Neural Information Processing Systems (Neur...

    2023

  6. [14]

    Y . Tu, W. Li, X. Song, K. Gong, L. Liu, Y . Qin, Y . Liu, and M. Liu. Using graph neural network to conduct supplier recommendation based on large-scale supply chain.International Journal of Production Research, pages 1–14, 2024

    2024

  7. [15]

    J. Kitchin. Cycles and trends in economic factors.Review of Economics and Statistics, 5(1):10–16, 1923

    1923

  8. [16]

    L. A. Metzler. The nature and stability of inventory cycles.Review of Economics and Statistics, 23(3):113–129, 1941

    1941

  9. [17]

    A. S. Blinder and L. J. Maccini. Taking stock: A critical assessment of recent research on inventories.Journal of Economic Perspectives, 5(1):73–96, 1991

    1991

  10. [18]

    Bils and J

    M. Bils and J. A. Kahn. What inventory behavior tells us about business cycles.American Economic Review, 90 (3):458–481, 2000

    2000

  11. [19]

    Hau L. Lee, V . Padmanabhan, and Seungjin Whang. Information distortion in a supply chain: The bullwhip effect. Management Science, 43(4):546–558, 1997. doi: 10.1287/mnsc.43.4.546. 20 MultiAgent-Inventory Inferring

    1997 doi

  12. [20]

    The bullwhip effect in supply networks.Management Science, 67(10):6153–6173, 2021

    Nikolay Osadchiy, William Schmidt, and Jing Wu. The bullwhip effect in supply networks.Management Science, 67(10):6153–6173, 2021. doi: 10.1287/mnsc.2020.3824

    2021 doi

  13. [21]

    Vertical contracts in a supply chain and the bullwhip effect.Management Science, 67(6): 3744–3756, 2021

    Zhan Qu and Horst Raff. Vertical contracts in a supply chain and the bullwhip effect.Management Science, 67(6): 3744–3756, 2021. doi: 10.1287/mnsc.2020.3630

    2021 doi

  14. [22]

    Strategic communication for capacity alignment with pricing in a supply chain.Management Science, 63(12):4366–4388, 2017

    Leon Yang Chu, Noam Shamir, and Hyoduk Shin. Strategic communication for capacity alignment with pricing in a supply chain.Management Science, 63(12):4366–4388, 2017. doi: 10.1287/mnsc.2016.2527

    2017 doi

  15. [23]

    The value of information design in supply chain management.Management Science, 71(8):6545–6558, 2025

    Ozan Candogan and Huseyin Gurkan. The value of information design in supply chain management.Management Science, 71(8):6545–6558, 2025. doi: 10.1287/mnsc.2023.03004

    2025 doi

  16. [24]

    A practical end-to-end inventory management model with deep learning.Management Science, 69(2):759–773, 2022

    Meng Qi, Yuanyuan Shi, Yongzhi Qi, Chenxin Ma, Rong Yuan, Di Wu, and Zuo-Jun Max Shen. A practical end-to-end inventory management model with deep learning.Management Science, 69(2):759–773, 2022. doi: 10.1287/mnsc.2022.4564

    2022 doi

  17. [25]

    Dynamic pricing with external information and inventory constraint.Management Science, 70(9):5985–6001, 2023

    Xiaocheng Li and Zeyu Zheng. Dynamic pricing with external information and inventory constraint.Management Science, 70(9):5985–6001, 2023. doi: 10.1287/mnsc.2023.4963

    2023 doi

  18. [26]

    Learning to order for inventory systems with lost sales and uncertain supplies.Management Science, 70(12):8631–8646, 2024

    Boxiao Chen, Jiashuo Jiang, Jiawei Zhang, and Zhengyuan Zhou. Learning to order for inventory systems with lost sales and uncertain supplies.Management Science, 70(12):8631–8646, 2024. doi: 10.1287/mnsc.2022.02476

    2024 doi

  19. [27]

    Kairen Zhang, Xiangyu Gao, Zhanyue Wang, and Sean X. Zhou. Sampling-based approximation for series inventory systems.Management Science, 71(10):8200–8217, 2025. doi: 10.1287/mnsc.2022.01876

    2025 doi

  20. [28]

    The effects of stock-based incentives on inventory management.Management Science, 68(7):5068–5086, 2022

    Qi Wu and Guoming Lai. The effects of stock-based incentives on inventory management.Management Science, 68(7):5068–5086, 2022. doi: 10.1287/mnsc.2021.4087

    2022 doi

  21. [29]

    Inventory as a financial instrument: Evidence from china’s metal industries.Management Science, 70(6):3645–3663, 2024

    Vernon Hsu and Jing Wu. Inventory as a financial instrument: Evidence from china’s metal industries.Management Science, 70(6):3645–3663, 2024. doi: 10.1287/mnsc.2023.4873

    2024 doi

  22. [30]

    Iancu, Nikolaos Trichakis, and Gerry Tsoukalas

    Dan A. Iancu, Nikolaos Trichakis, and Gerry Tsoukalas. Is operating flexibility harmful under debt?Management Science, 63(6):1730–1761, 2017. doi: 10.1287/mnsc.2015.2415

    2017 doi

  23. [31]

    Input specificity and the propagation of idiosyncratic shocks in production networks.Quarterly Journal of Economics, 131(3):1543–1592, 2016

    Jean-Noël Barrot and Julien Sauvagnat. Input specificity and the propagation of idiosyncratic shocks in production networks.Quarterly Journal of Economics, 131(3):1543–1592, 2016. doi: 10.1093/qje/qjw018

    2016 doi

  24. [32]

    Measuring the upstreamness of production and trade flows.American Economic Review, 102(3):412–416, 2012

    Pol Antràs, Davin Chor, Thibault Fally, and Russell Hillberry. Measuring the upstreamness of production and trade flows.American Economic Review, 102(3):412–416, 2012. doi: 10.1257/aer.102.3.412

    2012 doi

  25. [33]

    D. L. Iglehart and R. C. Morey. Inventory systems with imperfect asset information.Management Science, 18(8): B388–B394, 1972

    1972

  26. [34]

    DeHoratius and A

    N. DeHoratius and A. Raman. Inventory record inaccuracy: An empirical analysis.Management Science, 54(4): 627–641, 2008

    2008

  27. [35]

    Chehrazi

    N. Chehrazi. Inventory systems with record inaccuracy: Transaction errors vs. unobservable loss.Manufacturing & Service Operations Management, 27(4):1183–1204, 2025

    2025

  28. [36]

    Examining demand and supply-chain antecedents of inventory dynamics: Evidence from automotive industry.International Journal of Production Economics, 259:108838, 2023

    Chung-Yean Chiang, Zhuang Qian, Chia-Hung Chuang, Xiao Tang, and Chia-Ching Chou. Examining demand and supply-chain antecedents of inventory dynamics: Evidence from automotive industry.International Journal of Production Economics, 259:108838, 2023. doi: 10.1016/j.ijpe.2023.108838

    2023 doi

  29. [37]

    Z. Tan, B. Liu, and G. Yin. Asymmetric self-supervised graph neural networks. In2022 IEEE International Conference on Big Data (Big Data), pages 1369–1376. IEEE, 2022

    2022

  30. [38]

    The impact of supply chains on firm-level productivity.Management Science, 64(2):511–532, 2018

    Juan Carlos Serpa and Harish Krishnan. The impact of supply chains on firm-level productivity.Management Science, 64(2):511–532, 2018

    2018

  31. [39]

    When ignorance is not bliss: An empirical analysis of subtier supply network structure on firm risk.Management Science, 67(4):2029–2048, 2021

    Yimin Wang, Jing Li, Di Wu, and Ravi Anupindi. When ignorance is not bliss: An empirical analysis of subtier supply network structure on firm risk.Management Science, 67(4):2029–2048, 2021

  32. [40]

    Supply disruptions and optimal network structures

    Kostas Bimpikis, Ozan Candogan, and Shayan Ehsani. Supply disruptions and optimal network structures. Management Science, 65(12):5504–5517, 2019

    2019

  33. [41]

    Farias, Andrew A

    Vivek F. Farias, Andrew A. Li, and Tianyu Peng. Fixing inventory inaccuracies at scale.Manufacturing & Service Operations Management, 26(2):1102–1118, 2024

    2024

  34. [42]

    Xinyu Wang, Yong Lin, and Yina Shi. Linking industrial agglomeration and manufacturers inventory performance: The moderating role of firm size and enterprise status in the supply chain.Journal of Manufacturing Technology Management, 32(2):448–484, 2021. doi: 10.1108/JMTM-03-2020-0085

    2021 doi

  35. [43]

    Larson, Danko Turcic, and Fuqiang Zhang

    Chad R. Larson, Danko Turcic, and Fuqiang Zhang. An empirical investigation of dynamic ordering policies. Management Science, 61(9):2118–2138, 2015. 21 MultiAgent-Inventory Inferring

    2015

  36. [44]

    Martin A. Koschat. Store inventory can affect demand: Empirical evidence from magazine retailing.Journal of Retailing, 84(2):165–179, 2008. ISSN 0022-4359. doi: https://doi.org/10.1016/j.jretai.2008.04.003. URL https://www.sciencedirect.com/science/article/pii/S0022435908000146

    2008 doi

  37. [45]

    Baqaee and Emmanuel Farhi

    David R. Baqaee and Emmanuel Farhi. Supply and demand in disaggregated keynesian economies with an application to the covid-19 crisis.NBER Working Paper, (27152), 2020. doi: 10.3386/w27152

    2020 doi

  38. [46]

    Inventory productivity and stock returns in manufacturing networks

    Deepak Agrawal and Nikolay Osadchiy. Inventory productivity and stock returns in manufacturing networks. Manufacturing & Service Operations Management, 26(2):573–593, 2023. doi: 10.2139/ssrn.4166155

    2023 doi

  39. [47]

    Baqaee and Emmanuel Farhi

    David R. Baqaee and Emmanuel Farhi. Networks, barriers, and trade.Econometrica, 92(2):505–541, 2024. doi: 10.3982/ECTA19430. 22

    2024 doi

Pith tools

Reviewed July 14, 2026 · model on record in the stance chip above.