Pith. sign in

REVIEW 4 major objections 5 minor 47 references

FinPerMA: A Theory-Informed, Event-Grounded Personalized-Memory Benchmark for LLM Agents

T0 review · 4 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read FinPerMA: a benchmark that tests whether LLM agents update a user model after financial events, and shows they largely do not.

desk verdict FinPerMA's Post-Shock checkpoint is a real contribution to LLM memory evaluation, but its headline numbers should be read as provisional until the synthetic gold labels get a human spot-check. read the letter →

arxiv 2608.04095 v1 pith:GW7ID36R submitted 2026-08-04 cs.AI cs.CL

Ben Wang , Kang Zhou , Lifan Guo , Feng Chen , Chi Zhang This is my paper · ORCID
classification cs.AIcs.CL
keywords personalizedmemorybenchmarkLLMagentsevent-drivenpreferenceadaptationPost-ShockcheckpointImpactModelfinancialadvisorylongitudinalinvestorpersonassystemsevaluation
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 argues that current LLM agents are not actually maintaining a persistent, updated model of a user, and builds a benchmark to measure that claim. FinPerMA creates 276 synthetic investor personas with dated event timelines, generates dialogues with deterministic rules that specify how preferences should change after financial shocks, and freezes the corpus so every system faces the same trajectories. On 2,994 questions, no system—even one given the entire conversation history as context—exceeds about 47% overall accuracy or 39% on multiple-choice questions. The widest gaps appear right after a material event, when a memory system should revise a user's risk profile. The paper concludes that current memory architectures often preserve facts while losing the preference signals that personalization requires.

What carries the argument

The load-bearing object is the three-layer Impact Model. Layer 1 deterministically builds an ImpactConstraint for each persona-event pair: a four-way Behavioral Investor Type gates direction and forbidden biases, an event-type coefficient $c[\text{type}] \cdot \text{severity}/5$ sets magnitude, a Big-Five multiplier $1 + 0.30(P_N - 0.5) - 0.15(P_E - 0.5)$ scales it, an experience-history aggregation with weight exponent $\lambda = 1.3$ modulates it, and a symbolic memory-reference anchor (threshold $0.3$ on tag-Jaccard plus event-type match) supplies a reference value. Layer 2 narrates an event reaction under these hard constraints, and Layer 3 validates with automated checks and up to three retries. The Post-Shock checkpoint, posed immediately after a high-severity event, is what separates recall of the past from integration of new evidence into a persistent user model.

What would settle it

Take the frozen FinPerMA corpus and re-answer the Post-Shock questions with the true post-shock preference state (the Impact Model's intended updated values) injected directly into each memory system; if accuracy does not rise well above the reported ~39% MCQ, the benchmark is testing inference or recall rather than memory integration. A complementary check is to replace the rule-based gold labels with human-annotated or real-investor preference changes and see whether the relative ranking of retrieval versus structured memory reverses.

Watch

Extended reading notes

Core claim

The paper's central claim is that event-conditioned preference updating is a distinct capability that current LLM agents lack. On a frozen corpus of 2,994 questions from 276 personas, giving a model the entire conversation history as context lifts accuracy to only about 0.47 overall and about 39% on multiple choice, and the gap between full-context models and structured-memory systems grows from 8.0 points in-timeline to 13.0 points at the Post-Shock checkpoint. This is evidence that memory systems fail to integrate a material event into a persistent user model. The paper also claims that summary/profile memory preserves factual details at the expense of preference signals, so simple retrieval of raw dialogue recovers roughly 88% of the no-memory-to-full-context gap with about a tenth of the context tokens.

Load-bearing premise

The benchmark's gold labels stand or fall on the assumption that the hand-written Impact Model rules—event-type coefficients, Big-Five multiplier weights 0.30 and -0.15, experience exponent $\lambda = 1.3$, and memory threshold 0.3—are a valid stand-in for how real investors change preferences after events, so the generated dialogues and labels track true updating rather than generator artifacts.

Editorial extensions

If this is right

  • Full context alone does not solve personalization: even the strongest configuration leaves a wide accuracy gap, so the limiting factor is integration and reasoning, not context-window size.
  • Lightweight retrieval of verbatim dialogue recovers most of the benefit of full context at a fraction of the tokens, so memory systems should not summarise away the evidence that preference reasoning needs.
  • Structured/profile memory systems that consolidate facts are likely to keep failing on tasks that require inference or updating, because they discard preference signals.
  • The Post-Shock gap being wider than the in-timeline gap means memory systems have a specific, testable weakness in integrating consequential events into the user model.
  • The anti-typical-persona gap indicates that agents default to type-level stereotypes when individual evidence conflicts with a familiar profile, so personalization needs an override mechanism.

Reading between the lines

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

  • If the central claim holds, the Post-Shock protocol could be adapted as a general stress test for any long-horizon assistant (medical, legal, educational) whose advice should change after a material event.
  • A testable design implied by the paper: separate stable attributes from mutable preferences, timestamp preference deltas, and retrieve raw dialogue for preference-sensitive queries; this should close part of the gap without full context.
  • The deterministic Impact Model could double as a controllable data generator for counterfactual training, since it can produce trajectories with known preference updates on demand.
  • The absence of human validation of gold labels is the main external-validity threat; comparing the generated trajectories to real investor-panel or human-annotated preference changes would settle whether the Post-Shock results measure memory updating or generator artifacts.
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 introduces FinPerMA, a benchmark for evaluating LLM agents' personalized memory over longitudinal investor trajectories. The generation pipeline combines deterministic, theory-informed Impact Model rules (Eqs. 3–6), controlled LLM narration, and automated quality screening to produce a frozen corpus of 2,994 questions from 276 personas, including a Post-Shock checkpoint intended to test whether an agent integrates a material event into its persistent user model. Evaluation of seven frontier LLMs and up to seven memory configurations shows that no full-context configuration exceeds roughly 0.47 overall accuracy or 39% on multiple-choice questions, that retrieval-based memory approaches the full-context upper bound more efficiently than structured profile systems, and that the gap between full-context and mem0 widens from 8.0 in-timeline to 13.0 at the Post-Shock checkpoint. The paper concludes that memory systems often fail to update a user model after consequential events.

Significance. If the benchmark's gold labels are valid, FinPerMA addresses a real gap in personalized-memory evaluation by focusing on event-conditioned preference updating, and the Post-Shock checkpoint is a useful protocol contribution. The frozen corpus, deterministic rules, and release of code and prompts are concrete strengths that make the resource reproducible and amenable to future comparison. The empirical finding that all evaluated models remain far from saturated is striking and, if grounded in reliable labels, would be a valuable reference for memory-system design. However, the lack of human validation of the generated trajectories and labels, the single evaluation seed, and the unmatched token budgets across memory configurations temper the strength of the central conclusions; these issues are load-bearing rather than cosmetic.

major comments (4)
  1. [§2.4 (Eqs. 3–6) and Limitations] The gold labels are produced by hand-set Impact Model rules that the paper explicitly states are not an empirically estimated model of human behavior, and no human validation of the resulting trajectories, question answerability, or label correctness is reported. The abstract and §3.4 conclude that memory systems 'fail to integrate new evidence into a persistent user model,' but if the deterministic rules or LLM narration inject noise or if questions are not answerable from the provided evidence, then the observed ~0.47 ceiling and the Post-Shock gap would be artifacts of the generator rather than measurements of memory capability. The authors should report a human spot-check on a sample of questions (e.g., 100–200), including agreement on whether the intended preference change is inferable from the dialogue and whether the gold answer is unambiguously correct, and show that the main findings are unchanged on the human-validated subset.
  2. [§3.3 and Table 1, Panel B] The claim that the full_context−mem0 gap widens from 8.0 in-timeline to 13.0 at Post-Shock compares configurations with unmatched native token budgets: full_context uses 12.8k context tokens per query while mem0 uses 1.02k. The widening gap may simply reflect the smaller evidence budget available to mem0 at the Post-Shock checkpoint rather than a specific failure of memory updating. A token-controlled comparison, such as supplying the same number of retrieved tokens to each configuration or ablating the evidence budget within full_context, is needed to support the interpretation that the Post-Shock checkpoint isolates memory-updating failure as opposed to general evidence insufficiency.
  3. [§3.1 and Limitations] The memory-system comparison uses a single evaluation seed per configuration. Without repeated runs, confidence intervals, or significance tests, the reported differences among memory systems (e.g., Memobase vs Mem0, or the 8.0→13.0 gap) could be within sampling noise. Because the benchmark is frozen at 2,994 questions, one could bootstrap the question set or run multiple evaluation seeds with different random orders; either would provide variance estimates. The authors should report at least three seeds or bootstrap confidence intervals for the key comparisons in Table 1 and §3.3.
  4. [§3.1 (Metrics) and Table 1] Open-ended correctness is determined by majority vote of a three-way cross-vendor judge ensemble, with no validation of judge accuracy against human labels. Since 500 of the 2,994 questions are open-ended and the open-ended accuracy is a major driver of the memory-system ranking (retrieval's advantage on Open, and the MemFid scores used in attribution), judge bias could propagate to the central conclusions. The authors should report judge–human agreement on a human-annotated sample (e.g., Cohen's kappa or accuracy), or at least a secondary human evaluation on a subset of open-ended answers, to establish that the judge scores are reliable.
minor comments (5)
  1. [§3.1 and Table 1] Model names are inconsistent between the text and table (e.g., 'Qwen-3.8' in the table caption vs 'Qwen3.8' in the text, and 'Open' is used as a column header without a definition in the metrics paragraph). Please standardize naming and spell out each column abbreviation.
  2. [§2.6] The parenthesis '(novel)' in 'Post-Shock (novel)' is not explained; clarify whether 'novel' means the question is not seen during the dialogue or that the scenario is new.
  3. [Abstract and Table 1] The abstract reports 'approximately 0.47 overall accuracy' while Table 1 gives a maximum of 0.469; consider rounding consistently or reporting exact values in the abstract.
  4. [§3.1] The abbreviations 'PAS' and 'BIA' are used without expansion; define them at first use, perhaps as 'preference alignment score' and 'bias identification accuracy' if that is intended.
  5. [Figure 4] The radar and distractor-trap subfigures have small labels; consider increasing font sizes and adding a legend clearly mapping line styles to model names for readability.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the benchmark's headline numbers are empirical measurements on a frozen corpus, not consequences of the generation rules.

full rationale

FinPerMA's derivation chain is: (1) construct personas and event timelines; (2) compute ImpactConstraint via deterministic rules (Eqs. 3–6); (3) narrate dialogues under hard constraints; (4) generate questions; (5) evaluate held-out LLMs on the frozen v8gold corpus. The claims 'no full-context configuration exceeds approximately 0.47' and 'the full-context vs mem0 gap widens from 8.0 in-timeline to 13.0 at Post-Shock' are measurements on 2,994 frozen questions, not entailed by the Impact Model equations. The coefficients in Eqs. 3–6 are inputs selected by the authors, explicitly described as 'not an empirically estimated model of human behavior'; no parameter is fitted to the evaluated models' outputs and no evaluated model is defined in terms of the gold labels. The Post-Shock gap is an observed difference between full_context and mem0, not a construction identity. No load-bearing same-author citation or uniqueness theorem is used: prior work (Pompian 2012; Malmendier and Nagel 2011/2016; Bordalo et al. 2020) is external motivation, and the Mem0/MemOS/FinGPT baselines are evaluated as black boxes. The Limitations section honestly concedes the main non-circularity risk: 'FinPerMA uses synthetic personas and rule-guided preference changes, so benchmark accuracy does not establish performance with real investors' and notes that human review could test whether trajectories transfer. The absence of reported human validation of gold-label answerability is a validity threat—if labels leak or are unanswerable, the numbers could reflect generator artifacts—but that is a validity/external-evidence concern, not circularity, because the target claims are about LLM performance on this controlled corpus.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central benchmark construction rests on a set of hand-chosen coefficients, thresholds, and weights in the Impact Model, plus domain assumptions about behavioral-finance theory and LLM narration fidelity. There are no physical or conceptual entities like particles or forces; the Post-Shock checkpoint is an evaluation protocol design, not an entity. The absence of independent validation for these rules is the main burden on the benchmark's external validity.

free parameters (5)
  • Event-type shock coefficient c[E.type] = Not listed; lookup table over event types
    Eq. 3 sets base_mag as c[E.type] times severity/5; the values are chosen by hand to operationalize time-varying risk aversion, and the paper states it is not an empirically estimated structural coefficient.
  • Big-Five trait multiplier weights = 0.30 for neuroticism, -0.15 for extraversion
    Eq. 4 scales impact magnitude by personality; these weights are author-chosen with no sensitivity analysis or fitting reported.
  • Experience-weight exponent lambda = 1.3
    Eq. 5 uses age-dependent deterministic weights with exponent 1.3; the paper calls it a lifetime-experience heuristic, not an estimated parameter.
  • Belief modulation parameters = clip(1 - 0.2 * mu_hat, 0.6, 1.4)
    After Eq. 5, experienced returns modulate impact magnitude; the 0.2 coefficient and clipping bounds 0.6 and 1.4 are hand-chosen.
  • Memory similarity threshold and weights = 0.3 threshold, 0.8/0.2 tag/type weights
    Eq. 6 activates a memory when score >= 0.3, with 0.8 tag Jaccard and 0.2 event-type matching; these are author-defined, deliberately interpretable and auditable.
assumptions (4)
  • domain assumption Behavioral-finance theories (prospect theory, time-varying risk aversion, experience-based belief formation, memory-reference dependence) are valid grounds for generating realistic preference changes
    Section 2.4 cites Tversky and Kahneman 1992, Guiso et al. 2018, and others as motivation, but the operationalized rules are not empirically calibrated.
  • ad hoc to paper The Impact Model's deterministic rules do not need to be an estimated model of human behavior; the benchmark only requires internally consistent trajectories
    Section 2.4 states the scaffold is 'not an empirically estimated model of human behavior', which clarifies purpose but means the gold labels are author-defined.
  • domain assumption LLM narration with hard constraints and automated validation yields dialogues that faithfully represent the intended latent preference shifts without leaking the gold answer
    Stages 3 and 4 rely on the narrator and validator; the paper reports no human review or inter-annotator agreement on the generated trajectories.
  • domain assumption Public Chinese retail investor surveys are sufficient calibration targets for persona marginals
    Section 2.2 calibrates age, education, income, experience, and product participation to SZSE and AMAC reports; the rest use predefined priors.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FinPerMA: A Theory-Informed, Event-Grounded Personalized-Memory Benchmark for LLM Agents." pith.science (2026). https://pith.science/paper/GW7ID36R

@misc{pith2026260804095,
  author       = {Pith},
  title        = {Pith review of: FinPerMA: A Theory-Informed, Event-Grounded Personalized-Memory Benchmark for LLM Agents},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GW7ID36R}},
  note         = {Machine review of arXiv:2608.04095}
}
read the original abstract

Large language model (LLM) agents are increasingly used as personalized assistants in high-stakes domains such as financial advising, yet it remains unclear whether they can maintain and update an individualized user model over long horizons. Existing personalized-memory benchmarks primarily test factual retention or rely on weakly constrained model-generated trajectories, leaving event-driven preference adaptation underexplored. We introduce FinPerMA, an event-grounded benchmark that evaluates personalized memory against frozen longitudinal investor trajectories. Its generation pipeline combines deterministic, theory-informed impact rules, controlled LLM narration, and automated quality screening; a Post-Shock checkpoint isolates whether an agent has integrated a material event into its persistent user model. On 2,994 questions from 276 personas, seven frontier LLMs and up to seven memory configurations remain far from saturated: no full-context configuration exceeds approximately 0.47 overall accuracy or approximately 39% on multiple-choice questions. Attribution analysis shows that summary-based memory often preserves factual details while losing the preference signals needed for personalization; simple retrieval can therefore outperform purpose-built memory systems, with the gap widening after shocks.

Figures

Figures reproduced from arXiv: 2608.04095 by the authors.

Figure 1
Figure 1. The FinPerMA data-generation pipeline. Stage 1 samples personas from empirical distributions. Stage 2 constructs event timelines from real 2020–2026 events. Stage 3 applies the three-layer Impact Model ( [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The three-layer Impact Model. Layer 1 composes five deterministic sub-modules into an [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. A worked example of a single FinPerMA evaluation instance for persona P-0032. Panels 1–3 provide the persona profile, life-event timeline, and multi-session dialogue, respectively. The profile records traits and financial attributes. The timeline records macro, industry, and personal events through the Post-Shock checkpoint, while the dialogue reveals preferences implicitly. Panel 4 poses a six-option question with … view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Attribution radar under full_context. (a) Per-capability accuracy across the seven ability axes (zero-memory control questions removed); no single backbone leads on all axes. (b) Distractor-trap trigger rate by cognitive-failure mode, where each error option is pre-typ…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 17 canonical work pages

  1. [1]

    Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education

    Clancey, William J. Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education. Proceedings of the Eighth International Joint Conference on Artificial Intelligence (IJCAI-83)

  2. [2]

    Classification Problem Solving

    Clancey, William J. Classification Problem Solving. Proceedings of the Fourth National Conference on Artificial Intelligence

  3. [3]

    , title =

    Robinson, Arthur L. , title =. 1980 , doi =. https://science.sciencemag.org/content/208/4447/1019.full.pdf , journal =

    1980

  4. [4]

    New Ways to Make Microcircuits Smaller---Duplicate Entry

    Robinson, Arthur L. New Ways to Make Microcircuits Smaller---Duplicate Entry. Science

  5. [5]

    Clancey and Glenn Rennels , abstract =

    Diane Warner Hasling and William J. Clancey and Glenn Rennels , abstract =. Strategic explanations for a diagnostic consultation system , journal =. 1984 , issn =. doi:https://doi.org/10.1016/S0020-7373(84)80003-6 , url =

    doi:10.1016/s0020-7373(84)80003-6 1984

  6. [6]

    and Rennels, Glenn R

    Hasling, Diane Warner and Clancey, William J. and Rennels, Glenn R. and Test, Thomas. Strategic Explanations in Consultation---Duplicate. The International Journal of Man-Machine Studies

  7. [7]

    Poligon: A System for Parallel Problem Solving

    Rice, James. Poligon: A System for Parallel Problem Solving

  8. [8]

    Transfer of Rule-Based Expertise through a Tutorial Dialogue

    Clancey, William J. Transfer of Rule-Based Expertise through a Tutorial Dialogue

Show all 47 references
  1. [9]

    The Engineering of Qualitative Models

    Clancey, William J. The Engineering of Qualitative Models

  2. [10]

    2023 , eprint=

    Attention Is All You Need , author=. 2023 , eprint=

    2023

  3. [11]

    Pluto: The 'Other' Red Planet

    NASA. Pluto: The 'Other' Red Planet

  4. [12]

    Brown, Tom B. and Mann, Benjamin and Ryder, Nick and Subbiah, Melanie and Kaplan, Jared and Dhariwal, Prafulla and Neelakantan, Arvind and Shyam, Pranav and Sastry, Girish and Askell, Amanda and Agarwal, Sandhini and Herbert-Voss, Ariel and Krueger, Gretchen and Henighan, Tom ...

    2020 arXiv

  5. [13]

    Introducing Claude 4 , year =

  6. [14]

    and Cai, Carrie J

    Park, Joon Sung and O'Brien, Joseph C. and Cai, Carrie J. and Morris, Meredith Ringel and Liang, Percy and Bernstein, Michael S. , title =. Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology (UIST '23) , year =. doi:10.1145/3586183.3606763 ,...

  7. [15]

    Retrieval-Augmented Generation for Knowledge-Intensive

    Lewis, Patrick and Perez, Ethan and Piktus, Aleksandra and Petroni, Fabio and Karpukhin, Vladimir and Goyal, Naman and K. Retrieval-Augmented Generation for Knowledge-Intensive. Advances in Neural Information Processing Systems 33 (NeurIPS 2020) , year =. 2005.11401 , archivePrefix =

    2020 arXiv

  8. [16]

    and Zhang, Hao and Gonzalez, Joseph E

    Zheng, Lianmin and Chiang, Wei-Lin and Sheng, Ying and Zhuang, Siyuan and Wu, Zhanghao and Zhuang, Yonghao and Lin, Zi and Li, Zhuohan and Li, Dacheng and Xing, Eric P. and Zhang, Hao and Gonzalez, Joseph E. and Stoica, Ion , title =. Advances in Neural Information Processing ...

    2023 arXiv

  9. [17]

    , title =

    Jiang, Bowen and Xie, Zhuoqun and Hong, Yang and Ives, Zachary and Taylor, Camillo J. , title =. Proceedings of the Conference on Language Modeling (COLM 2025) , year =. 2504.14225 , archivePrefix =

    2025

  10. [18]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) , year =

    Maharana, Adyasha and Lee, Dong-Ho and Tulyakov, Sergey and Bansal, Mohit and Barbieri, Francesco and Fang, Yuwei , title =. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) , year =. 2402.17753 , archivePrefix =

    2024 arXiv

  11. [19]

    The Thirteenth International Conference on Learning Representations (ICLR 2025) , year =

    Wu, Di and Wang, Hongwei and Yu, Wenhao and Zhang, Yunsheng and Chang, Kai-Wei and Yu, Dong , title =. The Thirteenth International Conference on Learning Representations (ICLR 2025) , year =. 2410.10813 , archivePrefix =

    2025 arXiv

  12. [20]

    Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (ACL 2018) , year =

    Zhang, Saizheng and Dinan, Emily and Urbanek, Jack and Szlam, Arthur and Kiela, Douwe and Weston, Jason , title =. Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (ACL 2018) , year =. 1801.07243 , archivePrefix =

    2018 arXiv

  13. [21]

    and van Merri

    Weston, Jason and Bordes, Antoine and Chopra, Sumit and Rush, Alexander M. and van Merri. Towards. International Conference on Learning Representations (ICLR 2016) , year =. 1502.05698 , archivePrefix =

    2016 arXiv

  14. [22]

    and Stoica, Ion and Gonzalez, Joseph E

    Packer, Charles and Wooders, Sarah and Lin, Kevin and Fang, Vivian and Patil, Shishir G. and Stoica, Ion and Gonzalez, Joseph E. , title =. arXiv preprint , year =. 2310.08560 , archivePrefix =

    arXiv

  15. [23]

    Proceedings of the AAAI Conference on Artificial Intelligence (AAAI 2024) , year =

    Zhong, Wanjun and Guo, Lianghong and Gao, Qiqi and Ye, He and Wang, Yanlin , title =. Proceedings of the AAAI Conference on Artificial Intelligence (AAAI 2024) , year =. 2305.10250 , archivePrefix =

    2024 arXiv

  16. [24]

    arXiv preprint , year =

    Chhikara, Prateek and Khant, Dev and Aryan, Saket and Singh, Taranjeet and Yadav, Deshraj , title =. arXiv preprint , year =. 2504.19413 , archivePrefix =

    arXiv

  17. [25]

    arXiv preprint , year =

    Li, Zhiyu and Song, Shichao and Xi, Chenyang and Wang, Hanyu and Tang, Chen and Niu, Simin and Chen, Ding and Yang, Jiawei and Li, Chunyu and Yu, Qingchen and Zhao, Jihao and Wang, Yezhaohui and Liu, Peng and Lin, Zehao and Wang, Pengcheng and Huo, Jiahao and Chen, Tianyi and ...

    arXiv

  18. [26]

    Advances in Neural Information Processing Systems 38 (NeurIPS 2025) , year =

    Xu, Wujiang and Liang, Zujie and Mei, Kai and Gao, Hang and Tan, Juntao and Zhang, Yongfeng , title =. Advances in Neural Information Processing Systems 38 (NeurIPS 2025) , year =. 2502.12110 , archivePrefix =

    2025 arXiv

  19. [27]

    and Khashanah, Khaldoun , title =

    Yu, Yangyang and Li, Haohang and Chen, Zhi and Jiang, Yuechen and Li, Yang and Zhang, Denghui and Liu, Rong and Suchow, Jordan W. and Khashanah, Khaldoun , title =. Proceedings of the AAAI Spring Symposium Series (AAAI-SS 2024) / ICLR 2024 LLM Agents Workshop , year =. 2311.13...

    2024 arXiv

  20. [28]

    , title =

    Horton, John J. , title =. arXiv preprint , year =. 2301.07543 , archivePrefix =

  21. [29]

    and Busby, Ethan C

    Argyle, Lisa P. and Busby, Ethan C. and Fulda, Nancy and Gubler, Joshua R. and Rytting, Christopher and Wingate, David , title =. Political Analysis , year =

  22. [30]

    Leland , title =

    Bybee, J. Leland , title =. arXiv preprint , year =. 2308.16805 , archivePrefix =

    arXiv

  23. [31]

    arXiv preprint , year =

    Yang, Hongyang and Liu, Xiao-Yang and Wang, Christina Dan , title =. arXiv preprint , year =. 2306.06031 , archivePrefix =

  24. [32]

    Advances in Neural Information Processing Systems 37, Datasets and Benchmarks Track (NeurIPS 2024) , year =

    Xie, Qianqian and Han, Weiguang and Chen, Zhengyu and Xiang, Ruoyu and Zhang, Xiao and He, Yueru and Xiao, Mengxi and Li, Dong and Dai, Yongfu and Feng, Duanyu and Xu, Yijing and Kang, Haoqiang and Kuang, Ziyan and Yuan, Chenhan and Yang, Kailai and Luo, Zheheng and Zhang, Tia...

    2024 arXiv

  25. [33]

    Journal of Risk and Uncertainty , year =

    Tversky, Amos and Kahneman, Daniel , title =. Journal of Risk and Uncertainty , year =

  26. [34]

    Quarterly Journal of Economics , year =

    Barberis, Nicholas and Huang, Ming and Santos, Tano , title =. Quarterly Journal of Economics , year =

  27. [35]

    Quarterly Journal of Economics , year =

    Malmendier, Ulrike and Nagel, Stefan , title =. Quarterly Journal of Economics , year =

  28. [36]

    Quarterly Journal of Economics , year =

    Bordalo, Pedro and Gennaioli, Nicola and Shleifer, Andrei , title =. Quarterly Journal of Economics , year =

  29. [37]

    Journal of Finance , year =

    Bordalo, Pedro and Gennaioli, Nicola and Shleifer, Andrei , title =. Journal of Finance , year =

  30. [38]

    Journal of Financial Economics , year =

    Guiso, Luigi and Sapienza, Paola and Zingales, Luigi , title =. Journal of Financial Economics , year =

  31. [39]

    Evidence for Countercyclical Risk Aversion: An Experiment with Financial Professionals , journal =

    Cohn, Alain and Engelmann, Jan and Fehr, Ernst and Mar. Evidence for Countercyclical Risk Aversion: An Experiment with Financial Professionals , journal =. 2015 , volume =

    2015

  32. [40]

    Quarterly Journal of Economics , year =

    Enke, Benjamin and Graeber, Thomas , title =. Quarterly Journal of Economics , year =

  33. [41]

    A Model of Reference-Dependent Preferences , journal =

    K. A Model of Reference-Dependent Preferences , journal =. 2006 , volume =

    2006

  34. [42]

    and Cochrane, John H

    Campbell, John Y. and Cochrane, John H. , title =. Journal of Political Economy , year =

  35. [43]

    and Hirshleifer, David and Jiang, Danling and Lim, Sonya S

    Arkes, Hal R. and Hirshleifer, David and Jiang, Danling and Lim, Sonya S. , title =. Organizational Behavior and Human Decision Processes , year =

  36. [44]

    , title =

    Pompian, Michael M. , title =. 2012 , isbn =

    2012

  37. [45]

    and Newby, Rick and Sanghani, Jay , title =

    Durand, Robert B. and Newby, Rick and Sanghani, Jay , title =. Journal of Behavioral Finance , year =

  38. [46]

    2020 Individual Investor Survey Report (Summary) , year =

    2020

  39. [47]

    2021 , month = nov, note =

    National Public Fund Market Investor Survey Report (2020) , institution =. 2021 , month = nov, note =

    2020

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.