Observational Policy Ranking for SMB Financial Guidance from Multi-Action Accounting Logs

Ananya Pramod; Kamalika Das; Kratika Gupta; Padmaja Jonnalagedda; Ruocheng Guo; Shrutendra Harsola; Sricharan Kumar; Vignesh Subrahmaniam; Vikas Raturi; Xiang Gao

REVIEW 2 major objections 5 minor 25 references

Observational Policy Ranking for SMB Financial Guidance from Multi-Action Accounting Logs

T0 review · 2 major / 5 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read CAR-PL matches top policy for SMB growth guidance while spreading advice across the catalog

desk verdict Serious industrial policy-learning paper with a disciplined evaluation scaffold; the main caveat is an unvalidated off-support scoring probe, not an internal error. read the letter →

arxiv 2608.10050 v1 pith:GIAOIYMY submitted 2026-08-10 cs.LG

Shrutendra Harsola , Vignesh Subrahmaniam , Vikas Raturi , Kamalika Das , Xiang Gao , Kratika Gupta , Ruocheng Guo , Padmaja Jonnalagedda

show 2 more authors

Ananya Pramod Sricharan Kumar

This is my paper · ORCID

classification cs.LG

keywords observationalpolicyrankingmulti-actionlogsR-learnersupportshrinkagefinancialguidanceSMBmodel-assistedevaluation

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 whether useful financial guidance for small businesses can be learned from historical accounting logs, where business changes are self-selected and often happen together, instead of from randomized experiments. It introduces CAR-PL, a support-shrunk action-wise R-learner that learns one effect head per category while regularizing rare-category noise, and compares it with several baselines under a shared frozen scorer on company-disjoint held-out firms. The main claim is that CAR-PL ties with the T-Learner for the highest point estimates on Gross Profit and Revenue while selecting 33–34 categories, and that the contextual value model leads on Quick Ratio. The authors argue this supports objective-specific ranking of financial guidance from multi-action logs, and that recommendation concentration should be considered alongside aggregate scores.

What carries the argument

The central object is the Covariate-Adjusted Residual Policy Learning (CAR-PL) method: after cross-fitting a reward nuisance and per-category propensity models, it forms residuals for outcome and treatment, fits one weighted gradient-boosted effect head per category, and applies a support multiplier $\hat\delta^{\mathrm{shr}}_a(x) = \hat\delta_a(x)\, n_a/(n_a+\lambda_{\mathrm{sup}})$ before the argmax over 34 categories. This support shrinkage stabilizes selection of rare categories. The comparison itself rests on a shared model-assisted focal-action score $\hat s_i(a)$ that combines an outcome-model reference contrast with a bounded propensity-weighted residual correction, evaluated at the one-hot and zero treatment vectors.

What would settle it

Retrain the frozen outcome reference model with a different architecture (e.g., a transformer instead of the shared-trunk MLP) or with different pre-decision features, and re-score all policies on the same test rows; if the KPI-level point-estimate leaders change materially, the reported rankings are artifacts of the scoring model. Alternatively, run a prospective randomized study that assigns the top-scoring categories to firms and checks whether realized KPI changes match the model-assisted score ordering.

Watch

Extended reading notes

Core claim

The paper's central claim is that a support-shrunk action-wise R-learner (CAR-PL) matches the strongest baseline in aggregate model-assisted score on both growth KPIs while spreading recommendations across the catalog. On Gross Profit, CAR-PL has the highest point estimate (0.084); on Revenue, the T-Learner has the highest (0.085), with CAR-PL at 0.079; the paired differences are not statistically separated on either KPI. The contextual value model has the highest Quick Ratio point estimate (0.062). The paper also finds that CAR-PL and the T-Learner agree on only 12.8–14.2% of company-month states on the growth KPIs, showing that comparable aggregate scores can arise from markedly different state-to-category mappings. The robustness analyses (outcome-model-only scoring and alternative treatment references) retain the same KPI-level point-estimate leader or top pair.

Load-bearing premise

The comparison depends on the shared scorer evaluating categories through one-hot and zero treatment probes that never occur in the real multi-hot logs; if the outcome model extrapolates poorly to those off-support configurations, the KPI leaderboard could be an artifact of the measurement instrument rather than a property of the policies.

Editorial extensions

If this is right

If CAR-PL's support shrinkage is what enables broad coverage without losing aggregate score, then recommendation diversity becomes a reportable dimension in offline policy evaluation, not just the mean policy value.
The company-disjoint evaluation design, with firm-level clustered bootstraps for matched comparisons, provides a template for comparing policy families in settings where randomized recommendation labels are unavailable.
CAR-PL's parallel training of independent effect heads suggests the method scales to much larger category catalogs than 34, making it a practical option for industrial guidance systems.
The result that the zero-shot LLM concentrates on one category while outcome-fitted policies remain competitive supports the paper's claim that generic business priors are not a substitute for state-specific policy learning.
Objective-specific ranking matters: the KPI-dependent leaders mean that a single policy is not uniformly best, so deployment decisions should be made per financial objective.

Reading between the lines

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

If CAR-PL's broader catalog use is not merely a byproduct of the shrinkage multiplier but reflects genuinely different learned mappings, then downstream analyses of which states get which categories could reveal interpretable policy structure across business domains, a testable extension the paper does not pursue.
The stability of category rankings when replacing the all-zero treatment reference with the modal training co-action vector suggests that the shared scorer's outcome model is not strongly distorted by the off-support probes, but this robustness result is limited to the reported KPI-level leaders; per-state scoring instabilities could still exist.
A natural next step the paper leaves implicit is prospective validation: deploying the top-scoring policies in a randomized or A/B setting to measure whether the model-assisted scores predict actual KPI improvements from the recommended category.
The method's reliance on the shared scorer means that policy rankings could change if the outcome model were retrained with a different architecture or feature set; a sensitivity analysis across several outcome models would tell how much of the leaderboard is an artifact of the measurement instrument.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

What you should know: this paper earns its industrial claims. It sets up a clean prospective timeline, company-disjoint splits, firm-clustered bootstrap inference, a frozen policy-agnostic scorer, and a robustness grid. The result is a credible comparison of four policy families on multi-hot accounting logs, and CAR-PL's support-shrunk R-learning is genuinely new as a way to stabilize one-of-34 argmax selection. The headline pattern—CAR-PL highest on Gross Profit, T-Learner highest on Revenue, contextual value model highest on Quick Ratio, with no statistical separation between CAR-PL and T-Learner on either growth KPI—is consistent with the reported intervals and robustness checks. That is more than many applied papers deliver.

The real soft spot is the scorer's off-support probes. Eq. (9) evaluates the outcome model at one-hot and zero treatment vectors, but the training distribution is multi-hot. The paper never validates the model's extrapolation at those exact points. The residual correction is bounded and mostly inactive, and the outcome-model-only and modal-reference robustness checks do not touch the one-hot probe. The KPI rankings are therefore largely determined by that unvalidated contrast. To the authors' credit, they disclose this and name prospective validation as the next step. It is a missing calibration check, not a hidden error, but it means the claim that objective-specific ranking is feasible is conditional on the scoring model's extrapolation being trustworthy.

Other soft spots are proportionate and mostly external: proprietary data, withheld detector thresholds, and no code make independent replication impossible. The observational setting carries unmeasured-confounder risk, though the company-disjoint held-out evaluation at least tests generalization within the observed environment. There are several tuned constants—support multiplier, clipping bounds, CQL alpha—but the paper reports a sensitivity grid for Gross Profit and shows the outcome-model-only leaders are stable, which mitigates the concern.

Who this is for: people building decision support from operational logs, and researchers working on multi-action off-policy evaluation or policy learning with uneven support. It deserves a serious referee. The right outcome is likely a major revision that either validates the off-support scoring or softens the feasibility claim to match what the evidence supports.

Referee Report

2 major / 5 minor

Summary. This paper formulates the problem of ranking a single business-change category from multi-hot observational accounting logs as observational policy ranking. For each of three KPIs (Gross Profit, Revenue, Quick Ratio), it trains a separate policy: CAR-PL, an action-wise R-learner with support shrinkage; a T-Learner; a conservative contextual value model; plus zero-shot LLM, modal, and random references. All policies are evaluated on company-disjoint held-out firms with a shared frozen model-assisted score (Eq. 10). The paper reports KPI-specific point-estimate leaders, matched company-clustered confidence intervals, robustness to score components and treatment references, and selection-concentration analyses.

Significance. If the scoring instrument is accepted, the paper provides a credible industrial benchmark with careful evaluation design: company-disjoint splits, a frozen policy-agnostic scorer, firm-clustered bootstrapping, matched comparisons, and multiple robustness checks. CAR-PL's support-shrunk residual learning and broad catalog use are a useful methodological data point, and the paper is honest about the distinction between a model-assisted comparison score and an estimate of deployed policy value. However, because the ranking is mediated by a model-assisted score evaluated at standardized treatment probes, the significance of the central claim depends on validating those probes against observed multi-hot data.

major comments (2)

[Section 5.2, Eqs. (9)-(10)] The leading term of the shared score in Eq. (10) is the outcome-model contrast m_hat(w, u_a) - m_hat(w, 0) from Eq. (9), evaluated at the one-hot vector u_a and the zero vector. The training treatment distribution is multi-hot, and the paper does not report how often, if ever, each category appears as the sole active category; for categories with few or no exact-one-hot rows, the neural outcome model is extrapolating at precisely the probe that determines the score's first term. This is load-bearing because Table 7 shows the residual correction is saturated on only 2.6-12.9% of rows, and Table 9 shows outcome-model-only scoring retains the same leaders or top pair, so the rankings are effectively determined by these unvalidated contrasts. The existing modal-co-action robustness (Table 9) addresses only the reference side, not the one-hot side. Please add (i) exact-one-hot support counts per category, (ii) outcome-model calibration or rank correlation on exact-one-hot test rows if any exist, and (iii) a sensitivity analysis that replaces the one-hot probe with an observed co-action anchor, such as the average model prediction over training bundles containing category a. Without this, the headline KPI-level ranking claim rests on an untested extrapolation.
[Section 3.4 and Section 5.2] The target in Eq. (2) is defined as a focal-action contrast that averages over the observed co-action environment, but the score's first term compares a one-hot treatment vector to the zero vector. These are different estimands unless one assumes either that bundled co-actions are handled by the outcome model in a way that makes the one-hot contrast the relevant ranking instrument, or that the extrapolation to one-hot configurations is accurate. The paper should state precisely which estimand Eq. (10) is intended to score and why the one-hot-versus-zero contrast is the right ranking criterion for a policy that emits a single interpretable category from multi-hot logs. This clarification is needed for the central claim that the reported scores rank policies by objective-specific guidance quality.

minor comments (5)

[Abstract and Table 1] The abstract reports 85,078 observations from 7,505 firms, while Table 1 lists KPI-specific unique firm counts between 7,072 and 7,191; the text should clarify that 7,505 is the pre-filter corpus size and that KPI-specific filtering removes some firms.
[Table 6] The outcome-model diagnostics are computed on factual multi-hot test rows; please label them explicitly as factual diagnostics in the table and text, and distinguish them from any new diagnostics at the standardized probe configurations.
[Section 5.3] The clustered bootstrap resamples firms but does not address potential cross-firm calendar-time correlation from overlapping outcome windows; a sentence noting this limitation, or a two-way clustering check, would strengthen the inference section.
[Table 9] The zero/modal Spearman correlation is reported without specifying whether it is computed on training, validation, or test rows; please state the row population used for this robustness statistic.
[Section 7.6] The limitations paragraph mentions standardized treatment references but should explicitly flag the absence of an exact-one-hot support check as a limitation, rather than only listing the generic dependence on nuisance models.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the headline comparison is a held-out evaluation under a frozen, policy-agnostic scorer, and CAR-PL's support multiplier is selected on validation, not fitted to the reported test scores.

full rationale

The paper's central claim is an empirical comparison under a shared, frozen model-assisted scorer (Eq. 10), with company-disjoint held-out firms and test scores computed only after all model selection is frozen. Section 3.1 states: 'The test set is used only for the reported policy scores and confidence intervals.' CAR-PL's support multiplier is selected on validation (Section 4.3: 'Validation considers λsup ∈ {0, 1000, 2000, 4000} and yields λsup = 2000 for all three KPI policies'), then held fixed at test time; this is standard hyperparameter selection, not a fitted parameter renamed as a prediction. The shared scorer's outcome-model probes in Eq. (9) are one-hot and zero treatment vectors that are off the training support, and the paper explicitly labels the score as a standardized model-assisted comparison 'rather than as an estimate of deployed policy value' (Section 5.2); this is a measurement-instrument limitation, not a circular reduction. The robustness analyses in Table 9 show that the same KPI-level point-estimate leader or top pair is retained under outcome-model-only scoring and under a modal co-action treatment reference, so the headline ranking does not reduce to the residual correction. No load-bearing step is justified by a self-citation; the cited method papers (R-learning, CQL, off-policy evaluation) are external and are used as building blocks, not as the source of the reported findings. The paper's own Section 7.6 flags that prospective validation is the next step; that limitation is acknowledged rather than hidden. No equation is identical to its input by construction, and no fitted parameter is renamed as a prediction. Therefore no circular step meets the quoted-reduction threshold.

Assumptions & free parameters 7 free parameters · 7 assumptions · 0 invented entities

The central comparison depends on three families of inputs the paper does not pay for externally: causal identification assumptions (exchangeability, overlap, consistency) that are standard but unverifiable here; the ledger-detection layer whose precision is audited at 95% with unmeasured recall and proprietary thresholds; and the model-assisted scoring instrument whose off-support one-hot and zero probes are a design choice specific to this paper. The numeric constants (lambda_sup, clipping bounds, correction bound, winsorization, support threshold) are disclosed and mostly chosen on development data, so they are transparent but not independently grounded. No new entities are postulated; the action catalog is derived from the data.

free parameters (7)

support multiplier lambda_sup = 2000
Eq. (7) shrinks rare-category effect heads via n_a/(n_a+lambda_sup); selected from {0,1000,2000,4000} on the validation partition, yielding 2000 for all three KPIs. Directly shapes CAR-PL's argmax and its coverage profile.
catalog support threshold = 100 treated and 100 control observations per label
Section 3.2 retains 34 of 44 ledger labels with at least 100 treated and 100 control training rows; this outcome-independent rule defines the action space every policy is masked to.
propensity clipping bounds in scorer = 0.05 and 0.95
Section 5.1 clips scoring propensities to [0.05,0.95]; the treated-side clip rate is 4.8%, so the bounds are active and affect the residual correction term in Eq. (10).
residual correction bound = plus/minus 2
Section 5.2 clips the augmented residual term in Eq. (10) to [-2,2]; saturation is 2.6% to 12.9% across policies and KPIs in Table 7, so the bound is inactive on most rows but applies uniformly.
reward winsorization and cap = cap at 50; winsorize at training 5th and 95th percentiles
Section 3.3 transforms the raw KPI change before all modeling; these constants set the scale of every reported score and remove extreme outcomes.
CQL conservative weight alpha = 0.01
Eq. (3) balances the log-sum-exp penalty against regression; fixed rather than tuned per KPI for the contextual value baseline.
R-learner weight threshold and head settings = 1e-4 weight cutoff; 150 rounds; depth 4
Section 4.3 omits rows with weight below 1e-4 and fits depth-4 heads with 150 rounds; these settings control which observations contribute to each category's effect estimate.

assumptions (7)

domain assumption Conditional exchangeability given pre-decision state X: potential focal outcomes are independent of observed action choice given X
Section 3.4 identifies tau_a(x) under conditional exchangeability given pre-decision information; with self-selected actions this is unverifiable in the data and central to any causal reading of the scores.
domain assumption Positivity and overlap: each category has positive probability at each relevant state
Section 3.4 invokes positivity; the diagnostics in Table 6 (median treated ESS 58%, clip rate 4.8%) show overlap is only partial, so the assumption is approximately met at best.
domain assumption Consistency: the observed outcome under the observed bundle equals the potential outcome for that bundle
Standard causal assumption invoked in Section 3.4; with multi-hot bundles it requires well-defined potential outcomes for every treatment combination.
domain assumption Ledger-detected categories are valid proxies for business changes
Section 3.2 reports 95% audited precision on a 1,000-positive sample but no recall estimate; undetected changes would mislabel control months and bias contrasts.
ad hoc to paper The model-assisted score on standardized one-hot and zero probes is a valid ranking instrument
Section 5.2 defines the comparison metric via off-support probes in Eq. (9) and explicitly disclaims it as an estimate of deployed value; every headline ranking inherits this modeling choice.
domain assumption The focal-effect estimand averaging over the observed co-action environment is the right target for single-category ranking
Section 3.4 defines tau_a(x) to average over co-occurring categories; this makes the target task-appropriate but means the score is not the effect of cleanly implementing one change alone.
domain assumption No interference between firms (SUTVA)
Unstated but required for firm-level outcomes to be independent unit effects; plausible because firms are independent accounting entities.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Observational Policy Ranking for SMB Financial Guidance from Multi-Action Accounting Logs." pith.science (2026). https://pith.science/paper/GIAOIYMY

@misc{pith2026260810050,
  author       = {Pith},
  title        = {Pith review of: Observational Policy Ranking for SMB Financial Guidance from Multi-Action Accounting Logs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GIAOIYMY}},
  note         = {Machine review of arXiv:2608.10050}
}

read the original abstract

Small and medium-sized businesses need timely financial guidance, yet historical accounting logs record self-selected and often co-occurring business changes rather than randomized recommendations. We formulate this setting as observational policy ranking: from pre-decision financial information, a policy selects one of 34 ledger-derived business-change categories for a target financial KPI. Using 85,078 company-month observations from 7,505 firms, we introduce Covariate-Adjusted Residual Policy Learning (CAR-PL), an action-wise R-learner that operates directly on multi-hot logs and regularizes selection by observational support. We compare CAR-PL with an uplift T-Learner, a conservative contextual value model, a zero-shot LLM, and non-personalized references on company-disjoint held-out firms under a shared model-assisted scoring rule. CAR-PL has the highest Gross Profit point estimate (0.084), the T-Learner has the highest Revenue point estimate (0.085), and the contextual value model has the highest Quick Ratio point estimate (0.062). CAR-PL and the T-Learner are not statistically separated on either growth KPI in matched company-clustered comparisons, while CAR-PL selects 33-34 categories and produces less concentrated selections across the catalog. Outcome-model-only scoring retains the same KPI-level point-estimate leader or top pair, and category rankings remain similar when the all-zero treatment reference is replaced by the most common training co-action pattern. These findings support objective-specific ranking of SMB financial guidance from multi-action accounting logs.

Figures

Figures reproduced from arXiv: 2608.10050 by the authors.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 12 canonical work pages

[1]

The impact of consulting services on small and medium enterprises: Evidence from a randomized trial in mexico.Journal of Political Economy, 126 (2):635–687, 2018

Miriam Bruhn, Dean Karlan, and Antoinette Schoar. The impact of consulting services on small and medium enterprises: Evidence from a randomized trial in mexico.Journal of Political Economy, 126 (2):635–687, 2018. doi: 10.1086/696154

work page doi:10.1086/696154 2018
[2]

Policy learning with observational data.Econometrica, 89(1):133–161,

Susan Athey and Stefan Wager. Policy learning with observational data.Econometrica, 89(1):133–161,

work page
[3]

Offline multi-action policy learning: Generalization and optimization.Operations Research, 71(1):148– 183, 2023

Zhengyuan Zhou, Susan Athey, and Stefan Wager. Offline multi-action policy learning: Generalization and optimization.Operations Research, 71(1):148– 183, 2023. doi: 10.1287/opre.2022.2271

arXiv 2023
[4]

VanderWeele and Miguel A

Tyler J. VanderWeele and Miguel A. Hern´ an. Causal inference under multiple versions of treatment.Jour- nal of Causal Inference, 1(1):1–20, 2013. doi: 10. 1515/jci-2012-0002

work page 2013
[5]

K¨ unzel, Jasjeet S

S¨ oren R. K¨ unzel, Jasjeet S. Sekhon, Peter J. Bickel, and Bin Yu. Metalearners for estimating heteroge- neous treatment effects using machine learning.Pro- ceedings of the National Academy of Sciences, 116 (10):4156–4165, 2019. doi: 10.1073/pnas.1804597116

doi:10.1073/pnas.1804597116 2019
[6]

Quasi-oracle estima- tion of heterogeneous treatment effects.Biometrika, 108(2):299–319, 2021

Xinkun Nie and Stefan Wager. Quasi-oracle estima- tion of heterogeneous treatment effects.Biometrika, 108(2):299–319, 2021. doi: 10.1093/biomet/asaa076

doi:10.1093/biomet/asaa076 2021
[7]

Policy learning “without” overlap: Pessimism and generalized empirical bernstein’s inequality.The Annals of Statistics, 53(4):1483–1512, 2025

Ying Jin, Zhimei Ren, Zhuoran Yang, and Zhaoran Wang. Policy learning “without” overlap: Pessimism and generalized empirical bernstein’s inequality.The Annals of Statistics, 53(4):1483–1512, 2025. doi: 10. 1214/25-AOS2511

work page 2025
[8]

Off-policy deep reinforcement learning without ex- ploration

Scott Fujimoto, David Meger, and Doina Precup. Off-policy deep reinforcement learning without ex- ploration. InProceedings of the 36th International Conference on Machine Learning, volume 97 ofPro- ceedings of Machine Learning Research, pages 2052– 2062, 2019

work page 2019

Show all 25 references

[9]

Offline reinforcement learning: Tutorial, review, and perspectives on open problems.arXiv preprint arXiv:2005.01643, 2020

Sergey Levine, Aviral Kumar, George Tucker, and Justin Fu. Offline reinforcement learning: Tutorial, review, and perspectives on open problems.arXiv preprint arXiv:2005.01643, 2020

2005 arXiv
[10]

Conservative Q-learning for offline re- inforcement learning

Aviral Kumar, Aurick Zhou, George Tucker, and Sergey Levine. Conservative Q-learning for offline re- inforcement learning. InAdvances in Neural Informa- tion Processing Systems, volume 33, pages 1179–1191, 2020

2020
[11]

Khandani, Adlar J

Amir E. Khandani, Adlar J. Kim, and Andrew W. Lo. Consumer credit-risk models via machine-learning algorithms.Journal of Banking & Finance, 34(11): 2767–2787, 2010. doi: 10.1016/j.jbankfin.2010.06. 001

2010 doi
[12]

Predictably un- equal? the effects of machine learning on credit markets.The Journal of Finance, 77(1):5–47, 2022

Andreas Fuster, Paul Goldsmith-Pinkham, Tarun Ramadorai, and Ansgar Walther. Predictably un- equal? the effects of machine learning on credit markets.The Journal of Finance, 77(1):5–47, 2022. doi: 10.1111/jofi.13090. 9

2022 doi
[13]

Uncovering ChatGPT’s ca- pabilities in recommender systems.arXiv preprint arXiv:2305.02182, 2023

Sunhao Dai, Ninglu Shao, Haiyuan Zhao, Weijie Yu, Zihua Si, Chen Xu, Zhongxiang Sun, Xiao Zhang, and Jun Xu. Uncovering ChatGPT’s ca- pabilities in recommender systems.arXiv preprint arXiv:2305.02182, 2023

2023 arXiv
[14]

Christian Fieberg, Lars Hornuf, Maximilian Meiler, and David J. Streich. Using large language models for financial advice. CESifo Working Paper 11666, CESifo, 2025

2025
[15]

Qwen3 Embedding: Advancing text embedding and reranking through foundation models.arXiv preprint arXiv:2506.05176, 2025

Yanzhao Zhang, Mingxin Li, Dingkun Long, Xin Zhang, Huan Lin, Baosong Yang, Pengjun Xie, An Yang, Dayiheng Liu, Junyang Lin, Fei Huang, and Jingren Zhou. Qwen3 Embedding: Advancing text embedding and reranking through foundation models.arXiv preprint arXiv:2506.05176, 2025

2025 arXiv
[16]

Rosenbaum and Donald B

Paul R. Rosenbaum and Donald B. Rubin. The central role of the propensity score in observational studies for causal effects.Biometrika, 70(1):41–55,
[17]

d3rlpy: An offline deep reinforcement learning library.Journal of Ma- chine Learning Research, 23(315):1–20, 2022

Takuma Seno and Michita Imai. d3rlpy: An offline deep reinforcement learning library.Journal of Ma- chine Learning Research, 23(315):1–20, 2022

2022
[18]

LightGBM: A highly efficient gradient boosting decision tree

Guolin Ke, Qi Meng, Thomas Finley, Taifeng Wang, Wei Chen, Weidong Ma, Qiwei Ye, and Tie-Yan Liu. LightGBM: A highly efficient gradient boosting decision tree. InAdvances in Neural Information Processing Systems, volume 30, pages 3146–3154, 2017

2017
[19]

Decoupled weight decay regularization

Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. InInternational Conference on Learning Representations, 2019

2019
[20]

Robins, Andrea Rotnitzky, and Lue Ping Zhao

James M. Robins, Andrea Rotnitzky, and Lue Ping Zhao. Estimation of regression coefficients when some regressors are not always observed.Journal of the American Statistical Association, 89(427):846–866,
[21]

Dou- bly robust policy evaluation and learning

Miroslav Dud ´ ık, John Langford, and Lihong Li. Dou- bly robust policy evaluation and learning. InProceed- ings of the 28th International Conference on Machine Learning, pages 1097–1104, 2011

2011
[22]

Thomas and Emma Brunskill

Philip S. Thomas and Emma Brunskill. Data-efficient off-policy policy evaluation for reinforcement learning. InProceedings of the 33rd International Conference on Machine Learning, volume 48 ofProceedings of Machine Learning Research, pages 2139–2148, 2016. 10

2016
[1983]

doi: 10.1093/biomet/70.1.41

doi
[1994]

doi: 10.1080/01621459.1994.10476818

1994
[2021]

doi: 10.3982/ECTA15732

doi

Pith tools

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

[1] [1]

The impact of consulting services on small and medium enterprises: Evidence from a randomized trial in mexico.Journal of Political Economy, 126 (2):635–687, 2018

Miriam Bruhn, Dean Karlan, and Antoinette Schoar. The impact of consulting services on small and medium enterprises: Evidence from a randomized trial in mexico.Journal of Political Economy, 126 (2):635–687, 2018. doi: 10.1086/696154

work page doi:10.1086/696154 2018

[2] [2]

Policy learning with observational data.Econometrica, 89(1):133–161,

Susan Athey and Stefan Wager. Policy learning with observational data.Econometrica, 89(1):133–161,

work page

[3] [3]

Offline multi-action policy learning: Generalization and optimization.Operations Research, 71(1):148– 183, 2023

Zhengyuan Zhou, Susan Athey, and Stefan Wager. Offline multi-action policy learning: Generalization and optimization.Operations Research, 71(1):148– 183, 2023. doi: 10.1287/opre.2022.2271

arXiv 2023

[4] [4]

VanderWeele and Miguel A

Tyler J. VanderWeele and Miguel A. Hern´ an. Causal inference under multiple versions of treatment.Jour- nal of Causal Inference, 1(1):1–20, 2013. doi: 10. 1515/jci-2012-0002

work page 2013

[5] [5]

K¨ unzel, Jasjeet S

S¨ oren R. K¨ unzel, Jasjeet S. Sekhon, Peter J. Bickel, and Bin Yu. Metalearners for estimating heteroge- neous treatment effects using machine learning.Pro- ceedings of the National Academy of Sciences, 116 (10):4156–4165, 2019. doi: 10.1073/pnas.1804597116

doi:10.1073/pnas.1804597116 2019

[6] [6]

Quasi-oracle estima- tion of heterogeneous treatment effects.Biometrika, 108(2):299–319, 2021

Xinkun Nie and Stefan Wager. Quasi-oracle estima- tion of heterogeneous treatment effects.Biometrika, 108(2):299–319, 2021. doi: 10.1093/biomet/asaa076

doi:10.1093/biomet/asaa076 2021

[7] [7]

Policy learning “without” overlap: Pessimism and generalized empirical bernstein’s inequality.The Annals of Statistics, 53(4):1483–1512, 2025

Ying Jin, Zhimei Ren, Zhuoran Yang, and Zhaoran Wang. Policy learning “without” overlap: Pessimism and generalized empirical bernstein’s inequality.The Annals of Statistics, 53(4):1483–1512, 2025. doi: 10. 1214/25-AOS2511

work page 2025

[8] [8]

Off-policy deep reinforcement learning without ex- ploration

Scott Fujimoto, David Meger, and Doina Precup. Off-policy deep reinforcement learning without ex- ploration. InProceedings of the 36th International Conference on Machine Learning, volume 97 ofPro- ceedings of Machine Learning Research, pages 2052– 2062, 2019

work page 2019