Pith. sign in

REVIEW 3 major objections 7 minor 101 references

Experimental Designs for Multi-Item Multi-Period Inventory Control

T0 review · 3 major / 7 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read In multi-item inventory systems with a shared capacity, switchback A/B tests underestimate the true effect of raising base-stock levels, item-level randomization overestimates it, and a pairwise randomization over items and time sits…

desk verdict Clean bias-direction theory for inventory experiments, but the simulations violate the paper's own assumptions and the abstract overpromises real-data validation. read the letter →

arxiv 2501.11996 v4 pith:BMCFQL3V submitted 2025-01-21 stat.ME econ.EM

Xinqi Chen , Xingyu Bai , Zeyu Zheng , Nian Si This is my paper · ORCID
classification stat.MEecon.EM MSC 62K1590B05
keywords A/Btestingswitchbackexperimentsitem-levelrandomizationpairwiseinventorymanagementinterferenceglobaltreatmenteffectbase-stockpolicy
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

This paper asks how to run A/B tests on inventory-control policies when items share a warehouse and demand is lost when stock runs out. It shows that the two standard designs are systematically biased: switchback experiments, which flip all items between treatment and control by time period, underestimate the global treatment effect (GTE), while item-level randomization, which assigns each item to one arm for the whole horizon, overestimates it. The opposite bias directions arise because switchbacks suffer temporal carryover in inventory levels and item-level designs suffer cannibalization through the shared capacity constraint. The paper then proposes a pairwise design that randomizes both items and time, proves its bias lies between the two under a base-stock condition, and gives scenario-based guidance for choosing designs. If these results are right, practitioners can correct or avoid the bias rather than treating A/B estimates as neutral measurements.

What carries the argument

The argument runs on a telescoping profit identity. Because the firm orders up to $S_{n,t}$ each period and $X_{n,1}=0$, under Assumptions 2 or 3 the order is always placed, and the horizon profit of item $n$ telescopes to $\sum_{t=1}^H R_n^+(S_{n,t},D_{n,t})$, so the GTE is a sum of expected differences of the revenue function $R_n^+(s,d)=(r_n-c_n)(s-(s-d)^+)$. The bias formulas then separate into switchback-type terms involving $c_n[\mathbb{E}(S_{n,t}(1)-D_{n,t})^+-\mathbb{E}(S_{n,t}(0)-D_{n,t})^+]$, which are nonpositive by monotonicity, and item-level terms involving the conditional revenue differences $\mathbb{E}[R_n^+(S_{n,t}(W_t),D_{n,t})\mid W_{n,t}=1]-\mathbb{E}[R_n^+(S_{n,t}(1),D_{n,t})]$, which are nonnegative because partial treatment raises an item's scaled base-stock level above its global-treatment level. Pairwise randomization combines both terms; condition (10) controls the size of the temporal term so the middle estimate is ordered.

What would settle it

Set up the Section 4.1 stationary system with Normal demand and trace $E[\widehat{GTE}^{SW}]-GTE$ and $E[\widehat{GTE}^{IR}]-GTE$ under the paper's own parameter choices; then truncate the demand below the Assumption 2 bound and repeat. If the signs of either bias change when the truncation is removed, the direction-of-bias theorems are confined to their assumption region; if the signs persist, the assumption is not the active constraint.

Watch

Extended reading notes

Core claim

On the paper's own terms, in a periodic-review, multi-item, lost-sales inventory system with a warehouse capacity $B$ and base-stock policies scaled by $k_t=\min(1, B/\sum_m s_{m,t})$, the inverse-probability-weighting estimator of the global treatment effect is directionally biased. Under Assumptions 1 and 2, $\mathbb{E}[\widehat{GTE}^{SW}]\le GTE$ (Theorem 1); under Assumption 3, $\mathbb{E}[\widehat{GTE}^{IR}]\ge GTE$ (Theorem 2). For pairwise randomization with $W_{n,t}\sim \mathrm{Bernoulli}(p)$ i.i.d., the bias is the sum of an item-level term and a temporal carryover term; Theorem 3 shows $\mathbb{E}[\widehat{GTE}^{PR}]\le \mathbb{E}[\widehat{GTE}^{IR}]$, and if base-stock levels never exceed the newsvendor critical fractile $F_{n,t}^{-1}((r_n-c_n)/r_n)$, then $\mathbb{E}[\widehat{GTE}^{SW}]\le \mathbb{E}[\widehat{GTE}^{PR}]\le \mathbb{E}[\widehat{GTE}^{IR}]$. Staggered rollouts overestimate GTE in stationary environments (Theorem 4). The intended upshot is not that one design is always best: under tight capacity or understocking switchback is recommended, under loose capacity or overstocking item-level randomization is recommended, and pairwise randomization is the balanced middle option.

Load-bearing premise

The load-bearing premise is Assumption 2 (and its item-level analogue Assumption 3): in every period, the highest possible base-stock level this period minus the lowest possible next period must be no larger than the smallest possible demand, so an order is always placed; this fails for demand distributions with unbounded lower support such as the Normal distributions used in the paper's own simulations.

Editorial extensions

If this is right

  • A switchback experiment gives a conservative bound: if it shows a positive effect, the true global treatment effect is at least that large, so a positive switchback result is a safe signal to roll out the policy.
  • Item-level randomization can make an ineffective or harmful policy look good, because the shared capacity lets treatment items keep higher base-stock levels than they would under global treatment.
  • Pairwise randomization over items and time inherits both bias terms, and under condition (10) its estimate lies between the switchback and item-level estimates, giving a built-in bracket on the true GTE.
  • Staggered rollouts should be reserved for stationary settings; under non-stationary demand or drifting base-stock levels their bias is not even sign-determined.
  • The numerical studies indicate the practical recommendation: tight capacity or understocking points to switchback, loose capacity or overstocking points to item-level randomization, and intermediate cases point to pairwise randomization.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The two theorems together imply a bracketing strategy the authors do not spell out: run switchback and item-level randomization on the same system; if both assumptions hold, the true GTE lies between the two estimates, and their gap is a measure of interference severity.
  • Because Assumptions 2 and 3 require demand's essential infimum to dominate base-stock drops, the direction-of-bias results are not directly applicable to the Normal and other unbounded-lower-support demand models used in the simulations; a truncated or modified demand model would be the natural way to make theory and numerics consistent.
  • The same two-sided randomization idea could be transferred to other operational experiments with temporal and cross-unit interference, such as dynamic pricing with shared inventory or shelf-space allocation, where one would expect a similar middle-ground design to bracket the treatment effect.
  • Condition (10) is testable before an experiment: if historical demand data can estimate the quantile $F_{n,t}^{-1}((r_n-c_n)/r_n)$ for each item and period, practitioners can check whether pairwise randomization will indeed be bounded below by switchback.
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

3 major / 7 minor

Summary. The paper studies inverse-probability-weighted estimation of the global treatment effect (GTE) in a multi-item, multi-period lost-sales inventory system with a warehouse capacity constraint and proportionally scaled base-stock policies. It analyzes four experimental designs—switchback, item-level randomization, pairwise randomization, and staggered rollouts—and proves, under monotonicity and inventory-feasibility assumptions, that switchback experiments underestimate GTE, item-level randomization overestimates GTE, and pairwise randomization has bias between the two when base-stock levels are bounded by the newsvendor quantile. The paper also reports synthetic stochastic simulations and gives scenario-specific design recommendations. The proof strategy is a telescoping profit identity that holds when inventory entering each period never exceeds that period's scaled base-stock level.

Significance. If the theorems are correct, this is a useful contribution to the literature on causal inference in operations: it gives closed-form bias decompositions, identifies opposite bias directions for the two canonical designs, and proposes a pairwise design with a formal middle-bias guarantee. The derivations are transparent, do not fit parameters to data, and the bias formulas are interpretable in terms of order-cost carryover and capacity-driven base-stock scaling. The main limitations are that the numerical study does not operate under the theorems' stated assumptions, and the abstract advertises real-data experiments that are absent from the manuscript. These issues must be resolved before the empirical claims can be accepted, but the theoretical core appears defensible.

major comments (3)
  1. [§4.1; Assumptions 2 and 3] Section 4.1 sets D_{n,t} ~ k_t + A_n sin(2π(t+φ_n)/7) + N(µ_n, 1.5), whose essential infimum is −∞. Assumptions 2 and 3 require Dbar_{n,t} ≥ max_w S_{n,t}(w) − min_{w'} S_{n,t+1}(w'), so the stochastic simulations cannot satisfy the theorems' hypotheses. This is not a technicality: the appendix proves X_{n,t+1} = (S_{n,t}(W_t) − D_{n,t})_+ ≤ S_{n,t+1}(W_{t+1}) by induction exactly under this assumption, and when the assumption fails the identity Stilde_{n,t} = S_{n,t}(W_t) can fail, adding bias terms that need not have a fixed sign. The paper's own statement in Section 4.1 that the numerical study does not "specify that 1 holds" further indicates that even Assumption 1 is not imposed. Consequently Figure 5 does not verify Theorems 1–3; please either re-run the experiments under distributions and parameters satisfying Assumptions 1–3 and verify the lower-bound conditions, or explicitly frame the simulations as robustness checks and report the frequency of periods in which no order is placed.
  2. [Abstract; §4] The abstract states that "trace-driven experiments on real-world fresh-retail data show that the same mechanisms persist in realistic environments with stockout substitution," but the manuscript contains no trace-driven or fresh-retail experiments: Section 4 consists only of the synthetic simulations in Sections 4.1 and 4.2. This is a load-bearing empirical claim in the abstract and must either be implemented in the paper or removed.
  3. [§4.2; Theorem 3, Condition (10)] The Uniform[an, an+3] simulations in Section 4.2 avoid the unbounded-support problem, but the paper still does not verify Assumptions 2/3 or Condition (10), i.e., max_{w∈{0,1}^N} S_{n,t}(w) ≤ F_{n,t}^{-1}((r_n−c_n)/r_n). Without these checks, the middle-bias ordering E[GTE^SW] ≤ E[GTE^PR] ≤ E[GTE^IR] in Theorem 3 is not tested by Figure 6, and the Table 4 recommendations that rely on PR being a balanced compromise are not supported by theorem-consistent evidence. Please add explicit numerical verification of both conditions, or state and test a clearly labeled robustness version of the theorem.
minor comments (7)
  1. [§3.2] The sentence "By contrasting Theorems 2 and 3, we observe that switchback experiments and item-level randomized experiments exhibit biases in opposite directions" should refer to Theorems 1 and 2, since the comparison is between switchback and item-level randomization.
  2. [Appendix A.2.1] The displayed line "X_{n,2} = (S_{n,1}−D_{n,1})_+ ≤ (S_{n,1}−D_{n,1})_+ ≤ S_{n,2}" contains a redundant first inequality; it should explicitly apply Assumption 2 to obtain (S_{n,1}−D_{n,1})_+ ≤ max_w S_{n,1}(w) − Dbar_{n,1} ≤ min_{w'} S_{n,2}(w') ≤ S_{n,2}(W_2).
  3. [Appendix A.2.1 and A.2.3] The symbol "/upmodels" appears to be a corrupted independence symbol; please replace it with proper notation such as "⊥" or an explicit statement of independence.
  4. [Assumptions 2 and 3] Assumptions 2 and 3 use the same symbol D_{n,t} for the demand variable and for its essential infimum; please use a distinct notation, such as \underline{D}_{n,t}, in the displayed statements to avoid ambiguity.
  5. [§4.1] The sentence "we do not specify that 1 holds" should be completed to refer to Assumption 1 or equation (1), and the intended meaning should be clarified, as the current wording is ambiguous.
  6. [Tables 2 and 3] Several entries in the non-stationary rows, such as "φCn − 0.25 φn" and "0.8µn µn", are difficult to parse; please use unambiguous column entries or a legend.
  7. [§1.1] There is a typo in "SUTV A"; it should be "SUTVA".

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the bias theorems are derived algebraically from the stated model, estimator, and explicit assumptions, with no fit-to-data step and no load-bearing self-citation.

full rationale

The derivation chain is self-contained. The GTE and the IPW estimator are defined in Section 2.3 and Eq. (5), and Theorems 1-4 are obtained by substituting the inventory dynamics (2), the scaling rule (4), and the assignment distributions into E[GTE_hat]-GTE. Assumptions 1-3 and condition (10) are stated sufficient conditions used in the proofs; they are not imposed to match the conclusions. In particular, Assumptions 2 and 3 are used in the induction leading to Eq. (A.2) so that S-tilde_{n,t}=S_{n,t}(W_t), and condition (10) is used only to sign the PR-vs-SW comparison via the newsvendor marginal-profit property; neither quantity is fitted from data. No parameter is estimated from a subset and then 'predicted'; the paper contains no fitted values at all. The authors' own prior works (e.g., Si 2023, Weng 2024, Wu 2022, Simchi-Levi 2023) appear only in the literature review and are not cited as justification for any theorem or assumption. The numerical simulations' use of Normal demand, which violates the essential-infimum lower bound in Assumptions 2/3, is a genuine scope/validity concern about whether Figure 5 verifies the theorems, but it is a correctness domain issue rather than circular reasoning; the theorems themselves are not assumed to derive themselves. No self-definitional, fitted-input, or citation-forced step was found.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The theoretical results are derived under standard stochastic inventory assumptions plus the proportional scaling heuristic. No parameters are fitted to data; the simulation settings are chosen to illustrate different capacity and demand regimes. The central proofs rest on demand-support and order-placement assumptions that the numerical examples do not actually satisfy.

assumptions (5)
  • domain assumption Demand D_{n,t} has a positive essential infimum for all items and periods.
    Stated in Section 2.1: 'We assume D_{n,t} > 0'. Used implicitly in the proofs to keep the inventory dynamics non-degenerate and in Assumptions 2 and 3.
  • domain assumption Scaled base-stock levels are monotone: 0 <= S_{n,t}(0) <= S_{n,t}(1) (Assumption 1).
    Required for Theorem 1 to conclude that switchback estimates are biased downward; Lemma 1 gives sufficient conditions, but the assumption itself is taken as given.
  • domain assumption Between-period base-stock drops are bounded by the minimal demand (Assumption 2 for switchback, Assumption 3 for item-level and pairwise).
    This is the key premise that guarantees an order is always placed each period, so the telescoping profit identity holds and the bias formulas are exact. It is invoked in the proofs of Theorems 1, 2, 3, and 4.
  • domain assumption The inventory manager uses the proportional scaling heuristic (4) to enforce the capacity constraint.
    All base-stock levels in the analysis are computed as S_{n,t} = k_t s_{n,t} with k_t = min(1, B / sum s_m,t). The bias directions depend on this particular allocation rule.
  • domain assumption Treatment base-stock levels dominate control: s^C_{n,t} <= s^T_{n,t} for all n,t (equation (6)).
    Used to interpret the GTE as positive and to sign the bias terms; part of the 'somewhat uniform' treatment assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Experimental Designs for Multi-Item Multi-Period Inventory Control." pith.science (2026). https://pith.science/paper/BMCFQL3V

@misc{pith2026250111996,
  author       = {Pith},
  title        = {Pith review of: Experimental Designs for Multi-Item Multi-Period Inventory Control},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BMCFQL3V}},
  note         = {Machine review of arXiv:2501.11996}
}
read the original abstract

Randomized experiments, or A/B testing, are the gold standard for evaluating interventions, yet they remain underutilized in inventory management. This study addresses this gap by analyzing A/B testing strategies in multi-item, multi-period inventory systems with lost sales and capacity constraints. We examine two canonical experimental designs--switchback experiments and item-level randomization--and show that both suffer from systematic bias due to interference: temporal carryover in switchbacks and cannibalization across items under capacity constraints. Under mild conditions, we characterize the direction of this bias in different scenarios. Motivated by two-sided randomization, we propose a pairwise design over items and time and analyze its bias properties. Controlled stochastic simulations verify the theoretical predictions, and trace-driven experiments on real-world fresh-retail data show that the same mechanisms persist in realistic environments with stockout substitution.

Figures

Figures reproduced from arXiv: 2501.11996 by the authors.

Figure 1
Figure 1. An example of switchback experiment The following assumption states that the base-stock levels after scaling in the treatment group are always greater than those in the control group. 6 [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. An example of item-level randomized experiment [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. An example of pairwise randomized experiment [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: An example of staggered rollouts design estimator in staggered rollouts experiment is E[GT E ̂SR] − GT E = N ∑ n=1 E[ cn NHp(1 − p) (Sn,Hn (WHn ) − Dn,Hn ) + I(Hn ≠ H) + 1 NHp H ∑ t=Hn+1 (R + n (Sn,t(Wt), Dn,t) − R + n (Sn,t(1), Dn,t)) − 1 NH(1 − p) Hn ∑ t=1 (R + n (Sn…
Figure 5
Figure 5. Figure 5: A/B tests: impact of capacity constraint [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: A/B tests: impact of supply-demand relationship [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

101 extracted references · 58 canonical work pages

  1. [1]

    Abdel-Malek, L. L. and Areeratchakul, N. (2007). A quadratic programming approach to the multi-product newsvendor problem with side constraints. European Journal of Operational Research , 176(3):1607--1619

    2007

  2. [2]

    and Jia, R

    Agrawal, S. and Jia, R. (2019). Learning in structured mdps with convex cost functions: Improved regret bounds for inventory management. In Proceedings of the 2019 ACM Conference on Economics and Computation , pages 743--744

    2019

  3. [3]

    Araman, V. F. and Caldentey, R. (2009). Dynamic pricing for nonperishable products with demand learning. Operations research , 57(5):1169--1188

    2009

  4. [4]

    Aronow, P. M. and Samii, C. (2017). Estimating average causal effects under general interference, with application to a social network experiment. The Annals of Applied Statistics , pages 1912--1947

    2017

  5. [5]

    W., Masoero, L., McQueen, J., Richardson, T

    Bajari, P., Burdick, B., Imbens, G. W., Masoero, L., McQueen, J., Richardson, T. S., and Rosen, I. M. (2023). Experimental design in marketplaces. Statistical Science , 38(3):458--476

    2023

  6. [6]

    W., Soufiani, H

    Basse, G. W., Soufiani, H. A., and Lambert, D. (2016). Randomization and the pernicious effects of limited budgets on auction experiments. In Artificial Intelligence and Statistics , pages 1412--1420. PMLR

    2016

  7. [7]

    and Raouf, A

    Ben-Daya, M. and Raouf, A. (1993). On the constrained multi-item single-period inventory problem. International Journal of Operations & Production Management

    1993

  8. [8]

    and Kallus, N

    Bertsimas, D. and Kallus, N. (2020). From predictive to prescriptive analytics. Management Science , 66(3):1025--1044

    2020

Show all 101 references
  1. [9]

    and Muharremoglu, A

    Besbes, O. and Muharremoglu, A. (2013). On implications of demand censoring in the newsvendor problem. Management Science , 59(6):1407--1424

    2013

  2. [10]

    P., and Sridhar, R

    Beyer, D., Sethi, S. P., and Sridhar, R. (2001). Stochastic multiproduct inventory models with limited storage. Journal of Optimization Theory and Applications , 111:553--588

    2001

  3. [11]

    P., and Sridhar, R

    Beyer, D., Sethi, S. P., and Sridhar, R. (2002). Average-cost optimality of a base-stock policy for a multi-product inventory model with limited storage. In Decision & Control in Management Science: Essays in Honor of Alain Haurie , pages 241--260. Springer

    2002

  4. [12]

    and Coey, D

    Blake, T. and Coey, D. (2014). Why marketplace experimentation is harder than it seems: The role of test-control interference. In Proceedings of the fifteenth ACM conference on Economics and computation , pages 567--582

    2014

  5. [13]

    Bojinov, I., Simchi-Levi, D., and Zhao, J. (2023). Design and analysis of switchback experiments. Management Science , 69(7):3759--3777

    2023

  6. [14]

    Boyarsky, A., Namkoong, H., and Pouget-Abadie, J. (2023). Modeling interference using experiment roll-out. arXiv preprint arXiv:2305.10728

    2023 arXiv

  7. [15]

    Bright, I., Delarue, A., and Lobel, I. (2024). Reducing marketplace interference bias via shadow prices. Management Science

    2024

  8. [16]

    Bu, J., Simchi-Levi, D., and Wang, C. (2022). Context-based dynamic pricing with partially linear demand model. Advances in Neural Information Processing Systems , 35:23780--23791

    2022

  9. [17]

    Burnetas, A. N. and Smith, C. E. (2000). Adaptive ordering and pricing for perishable products. Operations Research , 48(3):436--443

    2000

  10. [18]

    Candogan, O., Chen, C., and Niazadeh, R. (2024). Correlated cluster-based randomized experiments: Robust variance minimization. Management Science , 70(6):4069--4086

    2024

  11. [19]

    Chawla, S., Hartline, J., and Nekipelov, D. (2016). A/b testing of auctions. In Proceedings of the 2016 ACM Conference on Economics and Computation , pages 19--20

    2016

  12. [20]

    Chen, B., Chao, X., and Shi, C. (2021). Nonparametric learning algorithms for joint pricing and inventory control with lost sales and censored demand. Mathematics of Operations Research , 46(2):726--756

    2021

  13. [21]

    Chen, B., Jiang, J., Zhang, J., and Zhou, Z. (2024a). Learning to order for inventory systems with lost sales and uncertain supplies. Management Science

    2024

  14. [22]

    Chen, B., Simchi-Levi, D., Wang, Y., and Zhou, Y. (2022). Dynamic pricing and inventory control with fixed ordering cost and incomplete demand information. Management Science , 68(8):5684--5703

    2022

  15. [23]

    and Plambeck, E

    Chen, L. and Plambeck, E. L. (2008). Dynamic inventory management with learning about the demand distribution and substitution probability. Manufacturing & Service Operations Management , 10(2):236--256

    2008

  16. [24]

    Chen, S., Simchi-Levi, D., and Wang, C. (2024b). Experimenting on markov decision processes with local treatments. arXiv preprint arXiv:2407.19618

    2024 arXiv

  17. [25]

    Chen, W., Shi, C., and Duenyas, I. (2020). Optimal learning algorithms for stochastic inventory systems with random capacities. Production and Operations Management , 29(7):1624--1649

    2020

  18. [26]

    Y., Shanthikumar, J

    Chu, L. Y., Shanthikumar, J. G., and Shen, Z.-J. M. (2008). Solving operational statistics via a bayesian analysis. Operations research letters , 36(1):110--116

    2008

  19. [27]

    Cochran, W., Autrey, K., and Cannon, C. (1941). A double change-over design for dairy cattle feeding experiments. Journal of Dairy Science , 24(11):937--951

    1941

  20. [28]

    DeCroix, G. A. and Arreola-Risa, A. (1998). Optimal production and inventory policy for multiple products under resource constraints. Management Science , 44(7):950--961

    1998

  21. [29]

    Dhaouadi, W., Johari, R., and Weintraub, G. Y. (2023). Price experimentation and interference in online platforms. arXiv preprint arXiv:2310.17165

    2023

  22. [30]

    T., and Rong, Y

    Ding, J., Huh, W. T., and Rong, Y. (2024). Feature-based inventory control with censored demand. Manufacturing & Service Operations Management , 26(3):1157--1172

    2024

  23. [31]

    Doudchenko, N., Zhang, M., Drynkin, E., Airoldi, E., Mirrokni, V., and Pouget-Abadie, J. (2020). Causal inference with bipartite designs. arXiv preprint arXiv:2010.02108

    2020 arXiv

  24. [32]

    Downs, B., Metters, R., and Semple, J. (2001). Managing inventory with multiple products, lags in delivery, resource constraints, and lost sales: A mathematical programming approach. Management Science , 47(3):464--479

    2001

  25. [33]

    predict, then optimize

    Elmachtoub, A. N. and Grigas, P. (2022). Smart “predict, then optimize”. Management Science , 68(1):9--26

    2022

  26. [34]

    Erlebacher, S. J. (2000). Optimal and heuristic solutions for the multi-item newsvendor problem with a single capacity constraint. Production and Operations Management , 9(3):303--318

    2000

  27. [35]

    Fan, X., Chen, B., and Zhou, Z. (2022). Sample complexity of policy learning for inventory control with censored demand. Available at SSRN 4178567

    2022

  28. [36]

    Farias, V., Li, A., Peng, T., and Zheng, A. (2022). Markovian interference in experiments. Advances in Neural Information Processing Systems , 35:535--549

    2022

  29. [37]

    Farias, V., Li, H., Peng, T., Ren, X., Zhang, H., and Zheng, A. (2023). Correcting for interference in experiments: A case study at douyin. In Proceedings of the 17th ACM Conference on Recommender Systems , pages 455--466

    2023

  30. [38]

    Federgruen, A., Guetta, D., Iyengar, G., and Liu, X. (2023a). Multi-item inventory systems with joint expected value and chance constraints: Asymptotically optimal heuristics. Available at SSRN 4595318

    2023

  31. [39]

    Federgruen, A., Guetta, D., Iyengar, G., and Liu, X. (2023b). Scalable approximately optimal policies for multi-item stochastic inventory problems. Available at SSRN 4595316

    2023

  32. [40]

    Fradkin, A. (2019). A simulation approach to designing digital matching platforms. Boston University Questrom School of Business Research Paper Forthcoming

    2019

  33. [41]

    W., Johari, R., and Rasouli, M

    Glynn, P. W., Johari, R., and Rasouli, M. (2020). Adaptive experimental design with temporal interference: A maximum likelihood approach. Advances in Neural Information Processing Systems , 33:15054--15064

    2020

  34. [42]

    Godfrey, G. A. and Powell, W. B. (2001). An adaptive, distribution-free algorithm for the newsvendor problem with censored demands, with applications to inventory and distribution. Management Science , 47(8):1101--1112

    2001

  35. [43]

    Guo, S., Shi, C., Yang, C., and Zacharias, C. (2024). An online mirror descent learning algorithm for multiproduct inventory systems. Available at SSRN 4806687

    2024

  36. [44]

    Han, K., Basse, G., and Bojinov, I. (2024). Population interference in panel experiments. Journal of Econometrics , 238(1):105565

    2024

  37. [45]

    Han, K., Li, S., Mao, J., and Wu, H. (2023). Detecting interference in online controlled experiments with increasing allocation. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages 661--672

    2023

  38. [46]

    Harshaw, C., S \"a vje, F., Eisenstat, D., Mirrokni, V., and Pouget-Abadie, J. (2023). Design and analysis of bipartite experiments under a linear exposure-response model. Electronic Journal of Statistics , 17(1):464--518

    2023

  39. [47]

    Holtz, D., Lobel, R., Liskovich, I., and Aral, S. (2020). Reducing interference bias in online marketplace pricing experiments. arXiv preprint arXiv:2004.12489

    2020 arXiv

  40. [48]

    and Wager, S

    Hu, Y. and Wager, S. (2022). Switchback experiments under geometric mixing. arXiv preprint arXiv:2209.00197

    2022

  41. [49]

    Huh, W. T. and Rusmevichientong, P. (2009). A nonparametric asymptotic analysis of inventory planning with censored demand. Mathematics of Operations Research , 34(1):103--123

    2009

  42. [50]

    and Veinott Jr, A

    Ignall, E. and Veinott Jr, A. F. (1969). Optimality of myopic inventory policies for several substitute products. Management Science , 15(5):284--304

    1969

  43. [51]

    Imbens, G. W. and Rubin, D. B. (2015). Causal inference in statistics, social, and biomedical sciences . Cambridge university press

    2015

  44. [52]

    S., and Volfovsky, A

    Jagadeesan, R., Pillai, N. S., and Volfovsky, A. (2020). Designs for estimating the treatment effect in networks with interference. The Annals of Statistics , 48(2):679--712

    2020

  45. [53]

    Jia, S., Kallus, N., and Yu, C. L. (2023). Clustered switchback experiments: Near-optimal rates under spatiotemporal interference. arXiv preprint arXiv:2312.15574

    2023 arXiv

  46. [54]

    Johari, R., Li, H., Liskovich, I., and Weintraub, G. Y. (2022). Experimental design in two-sided platforms: An analysis of bias. Management Science , 68(10):7069--7089

    2022

  47. [55]

    and Sautmann, A

    Kasy, M. and Sautmann, A. (2021). Adaptive treatment assignment in experiments for policy choice. Econometrica , 89(1):113--132

    2021

  48. [56]

    Keskin, N. B. and Zeevi, A. (2014). Dynamic pricing with an unknown demand model: Asymptotically optimal semi-myopic policies. Operations research , 62(5):1142--1167

    2014

  49. [57]

    J., Shapiro, A., and Homem-de Mello, T

    Kleywegt, A. J., Shapiro, A., and Homem-de Mello, T. (2002). The sample average approximation method for stochastic discrete optimization. SIAM Journal on optimization , 12(2):479--502

    2002

  50. [58]

    Lariviere, M. A. and Porteus, E. L. (1999). Stalking information: Bayesian inventory management with unobserved lost sales. Management Science , 45(3):346--363

    1999

  51. [59]

    and Lau, A

    Lau, H.-S. and Lau, A. H.-L. (1995). The multi-product multi-constraint newsboy problem: Applications, formulation and solution. Journal of Operations Management , 13(2):153--162

    1995

  52. [60]

    Leung, M. P. (2022). Rate-optimal cluster-randomized designs for spatial interference. The Annals of Statistics , 50(5):3064--3087

    2022

  53. [61]

    Levi, R., Perakis, G., and Uichanco, J. (2015). The data-driven newsvendor problem: New bounds and insights. Operations Research , 63(6):1294--1306

    2015

  54. [62]

    Li, H., Zhao, G., Johari, R., and Weintraub, G. Y. (2022). Interference, bias, and variance in two-sided marketplace experimentation: Guidance for platforms. In Proceedings of the ACM Web Conference 2022 , pages 182--192

    2022

  55. [63]

    Li, S., Johari, R., Kuang, X., and Wager, S. (2023). Experimenting under stochastic congestion. arXiv preprint arXiv:2302.12093

    2023

  56. [64]

    and Kroer, C

    Liao, L. and Kroer, C. (2023). Statistical inference and a/b testing for first-price pacing equilibria. In International Conference on Machine Learning , pages 20868--20905. PMLR

    2023

  57. [65]

    Liao, L., Kroer, C., Leonenkov, S., Schrijvers, O., Shi, L., Stier-Moses, N., and Zhang, C. (2024). Interference among first-price pacing equilibria: A bias and variance analysis. arXiv preprint arXiv:2402.07322

    2024 arXiv

  58. [66]

    T., Krishnan, H., and Uichanco, J

    Lin, M., Huh, W. T., Krishnan, H., and Uichanco, J. (2022). Data-driven newsvendor problem: Performance of the sample average approximation. Operations Research , 70(4):1996--2012

    2022

  59. [67]

    Liyanage, L. H. and Shanthikumar, J. G. (2005). A practical inventory control policy using operational statistics. Operations research letters , 33(4):341--348

    2005

  60. [68]

    Lyu, C., Zhang, H., and Xin, L. (2024a). Ucb-type learning algorithms with kaplan--meier estimator for lost-sales inventory models with lead times. Operations Research

    2024

  61. [69]

    Lyu, J., Xie, J., Yuan, S., and Zhou, Y. (2024b). A minibatch-sgd-based learning meta-policy for inventory systems with myopic optimal policy. arXiv preprint arXiv:2408.16181

    2024 arXiv

  62. [70]

    Masoero, L., Vijaykumar, S., Richardson, T., McQueen, J., Rosen, I., Burdick, B., Bajari, P., and Imbens, G. (2024). Multiple randomization designs: Estimation and inference with interference. arXiv preprint arXiv:2401.01264

    2024

  63. [71]

    and Schmidt, C

    Nahmias, S. and Schmidt, C. P. (1984). An efficient heuristic for the multi-item newsboy problem with a single constraint. Naval Research Logistics Quarterly , 31(3):463--474

    1984

  64. [72]

    Ni, T., Bojinov, I., and Zhao, J. (2023). Design of panel experiments with spatial and temporal interference. Available at SSRN 4466598

    2023

  65. [73]

    Niederhoff, J. A. (2007). Using separable programming to solve the multi-product multiple ex-ante constraint newsvendor problem and extensions. European Journal of Operational Research , 176(2):941--955

    2007

  66. [74]

    Pouget-Abadie, J., Aydin, K., Schudy, W., Brodersen, K., and Mirrokni, V. (2019). Variance reduction in bipartite experiments through correlation clustering. Advances in Neural Information Processing Systems , 32

    2019

  67. [75]

    Powell, W., Ruszczy \'n ski, A., and Topaloglu, H. (2004). Learning algorithms for separable approximations of discrete stochastic optimization problems. Mathematics of Operations Research , 29(4):814--836

    2004

  68. [76]

    and Russo, D

    Qin, C. and Russo, D. (2022). Adaptivity and confounding in multi-armed bandit experiments. arXiv preprint arXiv:2202.09036

    2022 arXiv

  69. [77]

    Qin, H., Simchi-Levi, D., and Zhu, R. (2023). Sailing through the dark: Provably sample-efficient inventory control. Available at SSRN 4652347

    2023

  70. [78]

    Shi, C., Chen, W., and Duenyas, I. (2016). Nonparametric data-driven algorithms for multiproduct inventory systems with censored demand. Operations Research , 64(2):362--370

    2016

  71. [79]

    Shi, C., Wang, X., Luo, S., Zhu, H., Ye, J., and Song, R. (2023). Dynamic causal effects evaluation in a/b testing with a reinforcement learning framework. Journal of the American Statistical Association , 118(543):2059--2071

    2023

  72. [80]

    and Bayati, M

    Shirani, S. and Bayati, M. (2024). Causal message-passing for experiments with unknown and general network interference. Proceedings of the National Academy of Sciences , 121(40):e2322232121

    2024

  73. [81]

    Si, N. (2023). Tackling interference induced by data training loops in a/b tests: A weighted training approach. arXiv preprint arXiv:2310.17496

    2023 arXiv

  74. [82]

    Simchi-Levi, D., Wang, C., and Zheng, Z. (2023). Non-stationary experimental design under linear trends. Advances in Neural Information Processing Systems , 36:32102--32116

    2023

  75. [83]

    Snyder, L. V. and Shen, Z.-J. M. (2019). Fundamentals of supply chain theory . John Wiley & Sons

    2019

  76. [84]

    Thomke, S. H. (2020). Experimentation works: The surprising power of business experiments . Harvard Business Press

    2020

  77. [85]

    J., Wang, L., Wang, R., and Yenipazarli, A

    Turken, N., Tan, Y., Vakharia, A. J., Wang, L., Wang, R., and Yenipazarli, A. (2012). The multi-product newsvendor problem: Review, extensions, and directions for future research. Handbook of Newsvendor Problems , pages 3--39

    2012

  78. [86]

    Ugander, J., Karrer, B., Backstrom, L., and Kleinberg, J. (2013). Graph cluster randomization: Network exposure to multiple universes. In Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining , pages 329--337

    2013

  79. [87]

    Veinott Jr, A. F. (1965). Optimal policy for a multi-product, dynamic, nonstationary inventory problem. Management science , 12(3):206--222

    1965

  80. [88]

    Wang, Y., Chen, B., and Simchi-Levi, D. (2021). Multimodal dynamic pricing. Management Science , 67(10):6136--6152

    2021

  81. [89]

    and Mersereau, A

    Wang, Z. and Mersereau, A. J. (2017). Bayesian inventory management with potential change-points in demand. Production and Operations Management , 26(2):341--359

    2017

  82. [90]

    Weng, C., Lei, X., and Si, N. (2024). Experimental design in one-sided matching platforms. Available at SSRN 4890353

    2024

  83. [91]

    Wu, Y., Zheng, Z., Zhang, G., Zhang, Z., and Wang, C. (2022). Non-stationary A/B tests. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages 2079--2089

    2022

  84. [92]

    Xie, Y., Ma, W., and Xin, L. (2024). Vc theory for inventory policies. arXiv preprint arXiv:2404.11509

    2024

  85. [93]

    Xiong, R., Athey, S., Bayati, M., and Imbens, G. (2024). Optimal experimental design for staggered rollouts. Management Science , 70(8):5317--5336

    2024

  86. [94]

    Xiong, R., Chin, A., and Taylor, S. (2023). Data-driven switchback design. preprint URL https://www. ruoxuanxiong. com/data-driven-switchback-design. pdf

    2023

  87. [95]

    and Huh, W

    Yang, C. and Huh, W. T. (2024). A nonparametric learning algorithm for a stochastic multi-echelon inventory problem. Production and Operations Management , 33(3):701--720

    2024

  88. [96]

    L., Airoldi, E

    Yu, C. L., Airoldi, E. M., Borgs, C., and Chayes, J. T. (2022). Estimating the total treatment effect in randomized experiments with unknown network structure. Proceedings of the National Academy of Sciences , 119(44):e2208975119

    2022

  89. [97]

    Yuan, H., Luo, Q., and Shi, C. (2021). Marrying stochastic gradient descent with bandits: Learning algorithms for inventory systems with fixed costs. Management Science , 67(10):6089--6115

    2021

  90. [98]

    Zhan, R., Han, S., Hu, Y., and Jiang, Z. (2024). Estimating treatment effects under recommender interference: A structured neural networks approach. arXiv preprint arXiv:2406.14380

    2024

  91. [99]

    and Du, S

    Zhang, B. and Du, S. (2010). Multi-product newsboy problem with limited capacity and outsourcing. European Journal of Operational Research , 202(1):107--113

    2010

  92. [100]

    Zhang, H., Chao, X., and Shi, C. (2020). Closing the gap: A learning algorithm for lost-sales inventory systems with lead times. Management Science , 66(5):1962--1980

    2020

  93. [101]

    Zhao, J. (2024). Experimental design for causal inference through an optimization lens. In Tutorials in Operations Research: Smarter Decisions for a Better World , pages 146--188. INFORMS

    2024

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.