PaperlayerView exact revision on arXiv

Source-linked AI summary

PAWS: Policy-driven Agentic World Simulation

Tiviatis Sim, Jia Hui Woon, Xinming Gao, Chen Gao, Fengbin Zhu, Zheng Huanhuan, Chua Tat Seng, Kenji Kawaguchi

arXiv:2609.28547v1cs.AIcs.CEcs.MAcs.SI

TL;DR

Financial multi-agent simulation datasets rarely align policy interventions, stakeholder responses, and market context with historical evidence. PAWS builds a dataset connecting verified policy episodes to news, structured actions, organizations, and daily market context. Validation and replay findings highlight both documented policy timelines and the challenge of detecting and timing rare stakeholder actions.

  • Problem

    Financial multi-agent simulation datasets rarely connect policy intent, stakeholder actions, and market context to temporally aligned historical evidence.

    Source: §1 · p. 1

  • Method

    PAWS links verified policy episodes to dated news, source-grounded actions represented by multi-layer event frames, normalized organizations, and daily market-return context.

    Source: §1 · p. 1

  • Results

    89.4% initial agreement on interaction-mode labels was achieved by independent AI and human reviewers, while replay results show that high accuracy can mask missed stakeholder actions in sparse panels.

    Source: §7 · p. 7

  • Takeaways & Limitations

    PAWS provides an auditable observational and replay resource for studying policy-response dynamics and action-outcome alignment, with evaluation emphasizing active-action recall, timing, and response-cascade fidelity.

    Source: §7 · p. 7

    PAWS is a curated observational and replay resource, not a causal estimate of policy effects.

    Source: §7 · p. 7

Abstract

from arXiv · show

Policy interventions propagate through public communication, institutional decisions, and stakeholder responses, yet datasets for financial multi-agent simulation rarely connect these processes to temporally aligned historical evidence. We introduce PAWS, a Policy-driven Agentic World Simulation dataset covering 36 verified U.S. financial and economic policy episodes, 12,727 policy-linked news records, and 65,291 source-grounded stakeholder actions. Each action is linked to its supporting news and represented by a multi-layer event frame capturing its interaction mode, financial-action family and subtype, semantic attributes, and conditional mappings to external taxonomies. Entities are resolved to normalized organizations, and actions are aligned with daily market-return context to support policy-agent simulation replay. On 2,522 stratified action samples, independent AI and human reviewers achieved 89.4% initial agreement on interaction mode, with disagreements subsequently adjudicated. Case studies of the 2008 short-selling ban and 2001 decimalization recover documented policy timelines and associated market patterns across both dense and sparse news settings. A replay study further shows that high accuracy can mask failure to detect rare stakeholder actions, identifying action timing and calibration as central challenges. PAWS provides an auditable substrate for evaluating agent influence, policy-response cascades, and action-outcome alignment in historically grounded financial simulations.

1 Introduction

PAWS addresses the lack of historically grounded datasets that align policy intent, stakeholder actions, and market context for financial multi-agent simulation. It contributes a structured dataset for studying policy-centered agent behavior and replaying policy environments.

1 Introduction

  • Classical agent-based finance models use numerical states and hand-coded rules, while financial language and trading-agent systems rarely align policy intent, stakeholder actions, and market context historically (Liu et al. 2023; Li et al. 2025; Liu et al. 2021).This leaves difficult questions about plausible simulated responses to real policy shocks and differences between intended effects and observed market outcomes.Source: §1 · p. 1
  • Financial-policy episodes, dated news, stakeholder actions, and market-return context are aligned in PAWS to support evaluation in temporally grounded policy environments.The dataset is intended to move beyond isolated news classification and price prediction.Source: §1 · p. 1
  • 36 verified episodes contain 12,727 policy-linked news rows and 65,291 extracted actions, each represented with a multi-layer event frame rather than a single label.Frames include interaction mode, financial-action family and constrained subtype, semantic attributes, and conditional crosswalks to external taxonomies.Source: §1 · p. 1
  • The paper presents dataset analysis, label and action-validity validation, case studies, and resources for replication and simulation studies.These contributions position PAWS as a cohesive resource connecting historical policy shocks, news evidence, stakeholder actions, and market returns.Source: §1 · p. 1

2 Related Works

Related work spans financial and LLM-based agent simulations, financial text and event resources, and research on policy shocks and market responses. The paper identifies a gap in resources that connect policy interventions, stakeholder actions, and market outcomes as auditable historical evidence.

2 Related Works

  • The related-work discussion groups prior research into financial and LLM-based simulations, financial text and event resources, and policy-shock and market-response research.These lines motivate PAWS as a bridge between qualitative policy analysis and quantitative market outcomes.Source: §2 · p. 2
  • Traditional agent-based finance models study systemic risk and policy failures but are text-blind, while modern market simulators and LLM agents support trading research without curated historical links among policy, stakeholder behavior, and market context (Lebaron 2006; Farmer and Foley 2009; Byrd, Hybinette, and Balch 2019; Liu et al. 2021; Sun et al. 2023; Park et al. 2023; Zhou et al. 2023; Liu et al. 2023; Li et al. 2025; Xiao et al. 2024).The reviewed systems include ABIDES, FinRL, TradeMaster, FinGPT, QuantAgents, and TradingAgents.Source: §2 · p. 2
  • Financial NLP datasets and broader actor-event streams support extraction and event organization, but are not curated around financial-policy interventions with policy motivation, stakeholder actions, and market outcomes together (Chen et al. 2022; Sharma et al. 2023; Zheng et al. 2019; Huang et al. 2024; Ward et al. 2013; Kwak and An 2016; Hu, Moon, and Ng 2024; Sim et al. 2026).Their typical units include documents, relations, event mentions, or actor-event streams rather than policy episodes with intervention windows.Source: §2 · p. 2
  • Event studies estimate market impacts of discrete economic events, and policy-uncertainty and monetary-policy databases show the importance of linking policy communication to financial responses (MacKinlay 1997; Baker, Bloom, and Davis 2016; Acosta et al. 2025).PAWS complements this work with temporally grounded evidence of stakeholder actions and downstream market context rather than a single treatment-effect estimate.Source: §2 · p. 2

3 Methodology

PAWS builds a policy-centered event resource by linking verified policy episodes to dated news, extracted and structured stakeholder actions, normalized organizations, and market context. The resulting daily action records and replay view provide a temporal substrate for multi-agent simulation.

3.1 Dataset Generation

  • Policy episodes are retained when they have an identifiable announcement or implementation date, financial-market relevance, and enough public documentation for external verification.The current snapshot contains 36 verified episodes and 301 policy-specific query keys; candidates are manually checked against finance research and official documentation.Source: §3.1 · p. 2

3.2 Derived Replay View

  • Daily market tables provide contemporaneous context rather than labels for action extraction, and the short-selling-ban replay treats each policy-agent-day tuple as a decision point.The replay compares majority, Markov, network-threshold, hybrid temporal-network, structured-classifier, and bounded GPT-4o agent models across no action, Communication, and Direct Intervention labels.Source: §3.1 · p. 2, §3.2 · p. 4

3.1 Dataset Generation

  • For each policy, date-bounded news queries cover its policy window; recovered articles retain provenance, and questionable or irrelevant records receive human review.When a policy end date is unavailable, a fixed post-announcement window is used; unrecoverable articles may be retained with sufficient metadata and flagged for inspection.Source: §3.1 · pp. 2–3

3 Methodology

  • PAWS’s construction pipeline moves from policy-centered news through stakeholder-action records and organization resolution to market alignment and replay-ready agent-day panels.The overall methodology selects policy episodes, retrieves dated news, extracts and coalesces actions, structures event frames, resolves entities, and links market context.Source: §3 · p. 2, Figure 1 · p. 3

3.1 Dataset Generation

  • Actions extracted from policy-window news are represented as event frames combining interaction mode, financial-action family and subtype, semantic attributes, and conditional taxonomy crosswalks.Extraction groups articles by canonical source; the process also uses semantic-role and predicate-argument tools to support action interpretation.Source: §3.1 · p. 3
  • Entity aliases are mapped to normalized organizations, while records sharing organization, date, and action type are merged into daily actions that preserve links to policies, news, and raw actions.Action classification uses AI verification followed by human validation of stratified samples, with disagreements reassessed for golden labels.Source: §3.1 · p. 3

4 Dataset Overview

PAWS covers 36 verified policy episodes with layered action representations and conditional crosswalks; organizational classifications had 99% observed accuracy after manual verification.

4 Dataset Overview

  • PAWS contains 36 verified policy episodes, 301 query keys, 12,727 news rows, and 65,291 action and action-frame rows.Figures 2 and 3 summarize policy timelines and the stakeholder-to-action distribution.Source: §4 · p. 4

4.1 Multi-Layer Event-Frame Actions

  • Each action is represented as a structured event frame separating realization, financial function, and fine-grained mechanism.The frame design also supports compatible views through external crosswalks.Source: §4.1 · p. 4

4 Dataset Overview

  • Figures 2–3 show per-policy actor-action proportions and stakeholder diversity relative to corpus size, including Action and ActionFrame counts on a logarithmic scale.Source: Figure 3 · p. 4, Figure 2 · p. 4

4.1 Multi-Layer Event-Frame Actions

  • The eight-valued interaction mode distinguishes how an event is performed or where it lies in its realization lifecycle.Its categories distinguish a promise to provide funding from funding already provided.Source: §4.1 · p. 4
  • The 21-valued financial-action family captures broad event functions, while 99 subtypes specify supported mechanisms and are constrained by family.Figure 4 compares within-policy action-type, interaction-mode, and family composition across 36 episodes; cells are separately normalized by representation layer.Source: §4.1 · pp. 4–5, Figure 4 · p. 5
  • Six external crosswalk systems provide eligible ontology-specific views; not_applicable marks structural ineligibility rather than low confidence.The views cover political interactions, policy tools, general event types, monetary-policy classes, macroprudential tools, and crisis responses.Source: §4.1 · p. 5

4.2 Validation Results

  • 99% observed accuracy across 4,530 organization strings; the two incorrect actor assignments were corrected.Manual verification checked organizational classifications.Source: §4.2 · p. 5
  • Action validation reports reviewer agreement and human adjudication of conflicts, while deterministic-rule violations requiring auto-repair are tabulated.The supplied table captions identify interaction-mode agreement and deterministic database corrections, but do not give their values.Source: §4.2 · p. 5, Table 1 · p. 5, Table 2 · p. 5

5 Case Studies

The two case studies test PAWS against both a dense crisis-period intervention and a sparsely covered policy with a longer timeline. They recover dated actions and associated market patterns across these contrasting settings.

5 Case Studies

  • The short-selling ban and decimalization case studies recover complex action dynamics in a dense news window and dated actions in sparse coverage.The short-selling ban had 62 news articles and 639 actions; decimalization had 21 articles and 191 actions.Source: §5 · pp. 5–6
  • The actor distribution differs by policy: Policy 5 peaks among banks, insurance, and brokerages, while Policy 10 has no non-US-country or global-body actors.Both policies have a high proportion of United States official bodies.Source: Figure 5 · p. 6, §5 · p. 6

5.1 Policy 5: Short-selling ban

  • Policy 5’s timeline records the UK and US bans, additions to the banned-stock list, the extension, and the US ban’s end.The initial bans are reflected in 17 actions from September 18–19, 2008.Source: §5.1 · p. 6
  • Bid-ask spreads increased during and after the ban, with the widening persisting after it ended.The estimate uses daily high and low prices for 11 representative finance stocks.Source: Figure 6 · p. 6, §5.1 · p. 6
  • Finance-sector returns were worse than non-finance returns as the market plunged; Morgan Stanley’s lobbying was followed by retaliatory short-selling after the ban lifted.Figure 8 aligns large Morgan Stanley stock-price movements with dates extracted from news.Source: Figure 7 · p. 6, §5.1 · p. 6, Figure 8 · p. 7

5.2 Policy 10: Decimalization

  • For decimalization, PAWS reflects the SEC’s phase-in order, NYSE and Nasdaq implementation dates, and reported delays with the correct dates and classifications.The dataset also captures suspicions about insider profits and explanations of decimal pricing’s benefits to investors.Source: §5.2 · p. 7

6 Simulation Study

The replay predicts stakeholder actions from a policy-agent-day panel. Results show that aggregate accuracy can obscure poor recovery of rare active responses, with models trading off active recall and active-F1.

6 Simulation Study

  • The study treats each selected stakeholder and policy-agent-day as a decision point for predicting stakeholder actions.The panel’s target is one of no action, Communication, or Direct Intervention.Source: §6 · p. 7
  • GPT-4o reached .824 active recall but .122 active-F1, while the structured classifier reached .069 active-F1 and .176 active recall.The structured classifier has the best active-F1 in the full-process replay; GPT-4o has the highest active recall but over-triggers relative to the observed action rate.Source: Figure 9 · p. 7, §6 · p. 7
  • No-action diffusion baselines reached .939 accuracy but zero active-F1, despite an observed active rate of only 8.7%.They therefore failed to recover active stakeholder responses in the sparse policy-agent-day panel.Source: §6 · p. 7

7 Conclusions and Future Work

PAWS connects verified policy episodes to dated news, source-grounded stakeholder actions, structured event frames, normalized organizations, and daily market context. Validation and case studies support this design while exposing replay-evaluation challenges and a boundary on causal interpretation.

7 Conclusions and Future Work

  • PAWS links verified financial-policy episodes with dated news, source-grounded actions, normalized organizations, structured event frames, and daily market context.Source: §7 · p. 7
  • Independent AI and human reviewers reached 89.4% initial agreement on interaction-mode labels before disagreements were adjudicated.Case studies recover documented timelines and associated market patterns in both dense- and sparse-news settings.Source: §7 · p. 7
  • High replay accuracy can result from predicting no action while missing stakeholder responses, so evaluation should emphasize active-action recall, timing, and response-cascade fidelity.Source: §7 · p. 7
  • PAWS is a curated observational and replay resource, not a causal estimate of policy effects.Future work should expand policy and jurisdiction coverage, strengthen human validation, improve agent calibration, and add higher-fidelity market mechanisms and counterfactual evaluation.Source: §7 · p. 7

B.14 19: U.S. Implementation of Basel III

The policy episodes span monetary, regulatory, fiscal, and trade interventions, with documented market responses that sometimes differ from policy intent. The section also motivates distinguishing an event’s financial function from how it is realized and whether it is intended or observed.

B.15 15: Quantitative Easing 3 (Open-Ended LSAP)

  • Other monetary episodes include QE3 purchases intended to support housing and asset prices, and the Evans Rule’s shift from calendar guidance to explicit economic thresholds.The QE3 passage reports higher global equity prices, lower VIX, and emerging-market capital inflows (Eichengreen and Gupta 2015); the Evans Rule passage describes thresholds of 6.5% unemployment and 2.5% inflation (Justiniano et al. 2012).Source: §B.15 · p. 12, §B.16 · p. 12

B.17 24: Unconventional Monetary Policy Tapering Announcement (Taper Tantrum)

  • The 2013 tapering announcement was reported as a monetary shock that increased risk compensation and triggered capital outflows from emerging markets.The cited research also reports disproportionate declines in emerging-market equities and currencies relative to physical capital flows (Chari, Dilts Stedman, and Lundblad 2021).Source: §B.17 · pp. 12–13

B.19 26: The Volcker Rule (Dodd-Frank Act Section 619)

  • The Volcker Rule was associated with reduced corporate-bond liquidity during stress, while its later revisions simplified compliance and were intended to reverse illiquidity.Bao, O’Hara, and (Alex) Zhou (2018) report reduced market-making and increased illiquidity when bonds were downgraded; the passage describes the 2019/2020 revisions as enacted to reverse this illiquidity.Source: §B.19 · p. 13, §B.31 · p. 15

B.26 33: U.S.-China Trade War Tariff (1)

  • Trade-war tariff announcements had differing reported market effects: the first and third tranches coincided with losses, while the second had little net market impact.The reported effects include a -3.39% seven-day cumulative return for the first announcement and a -0.60% market effect for the second; the third coincided with a -1.62% broader-market return.Source: §B.26 · p. 14, §B.28 · p. 14, §B.29 · p. 14

B.32 39: Federal Reserve Emergency Rate Cuts and Corporate Credit Facilities (PMCCF/SMCCF)

  • Emergency corporate-credit facilities were associated with sharply lower eligible-bond spreads after announcement and further spread declines after physical purchases.Gilchrist et al. (2024) report a 70-basis-point decline within 10 days of announcement and an additional 11-basis-point decline after purchases; Boyarchenko, Kovner, and Shachar (2022) report over 90 basis points within three days.Source: §B.32 · p. 15

B.33 40: Coronavirus Aid, Relief, and Economic Security (CARES) Act

  • The CARES Act stimulus checks coincided with abnormal returns in retail-concentrated portfolios, whereas the third stimulus round had no significant positive portfolio-return effect.Greenwood, Laarits, and Wurgler (2022) report 5.62% first-day abnormal returns and 10.73% cumulative abnormal returns over three days for the CARES Act episode; the American Rescue Plan passage reports no significant positive effect.Source: §B.33 · p. 15, §B.34 · p. 15

C Alternative Policy Identification LLM

D.1 Event-Frame Forms

  • The event frame separates interaction mode from financial-action family and subtype, so a plan, authorization, and completed transfer can share a function but differ in realization and evidential status.Conditional external crosswalks provide compatible views when applicable; not_applicable means an ontology does not structurally license the action, not that the action is missing or invalid.Source: §D.1 · p. 16

D.3 Dataset Statistics

Dataset statistics describe entity normalization and the distributions of action families, subtypes, news volume, and organizational policy coverage.

Entity

  • 7,810 of 13,720 unique nonblank entity strings (56.9%) were assigned to an organization, yielding 4,317 canonical labels, of which 2,578 (59.7%) linked to distinct Wikidata QIDs.The accompanying table states that entity percentages use unique entity strings as the denominator, except Wikidata coverage, which uses canonical labels.Source: §D.3 · p. 16, Table 4 · p. 17

Event-Frame Distributions

  • Figures 10 and 11 show financial-action family composition by policy and the overall distribution of action subtypes, respectively.Each figure displays only the most frequent categories and pools the remainder; numeric codes map to a category key.Source: §D.3 · p. 16

News Information

  • Figure 12 presents the distributions of daily origin-news counts for each keyword, organized by policy, showing how origin-news counts are distributed across keywords under the different policies.Source: §D.3 · p. 16

Stakeholder Information

  • Figure 13 ranks organizations by distinct policy coverage and reports the corresponding number of Action rows, excluding generic actors, countries, markets, and indices.Source: §D.3 · p. 16

E Detailed Case Study Data

The case-study appendix lists action records associated with the short-selling ban and decimalization, alongside diagnostics and a comparison of resources relevant to policy-driven simulation.

E Detailed Case Study Data

  • The short-selling-ban action lists distinguish U.S. and U.K. announcements and implementations, positive-return examples, blame and opposition, and Morgan Stanley retaliatory shorting.The decimalization timeline records the phase-in announcement, delays, exchange completion milestones, and U.S. completion.Source: E Detailed Case Study Data · p. 17
  • Decimalization action records separately identify suspected profits from fractional pricing and reported investor benefits from decimal pricing.Source: E Detailed Case Study Data · p. 17

F Additional Simulation Diagnostics

G Additional Comparison to Related Work

  • The related-work comparison evaluates whether simulators, financial NLP datasets, event-stream resources, and event-study databases preserve policy evidence, intent, stakeholder actions, entity relevance, and realized market context.Source: G Additional Comparison to Related Work · p. 17

H Detailed Research Questions and Use Cases

PAWS is framed as evidence for studying influence between agents, alignment between actions and outcomes, and expected versus unexpected policy effects, while leaving causal identification unresolved. The appendix also specifies a larger short-selling-ban replay and possible extensions to its modular design.

H Detailed Research Questions and Use Cases

  • PAWS does not itself solve causal identification in multi-agent systems, but supplies temporally aligned evidence for defining and evaluating related questions.Source: H Detailed Research Questions and Use… · p. 17
  • Linked dated stakeholder actions and organizations support multi-turn analysis of which agents influence others and when.The passage presents selective influence among agents at particular timestamps as a future research direction.Source: H Detailed Research Questions and Use… · p. 17
  • Action types, actor identities, dates, and policy context support tests of whether simulated outcomes align with agents’ actions.The passage identifies agent memory, interaction topology, policy conditioning, and temporal update rules as design choices for such research.Source: H Detailed Research Questions and Use… · p. 17
  • Table 5 compares related work on whether it jointly preserves policy intent, stakeholder behavior, entity relevance, and realized market context for policy-driven simulation.Source: Table 5 · p. 18
  • Documented policy intentions and observed market effects support comparing intended goals with stakeholder reactions and market context, including research on counterfactuals and second-order effects.Source: H Detailed Research Questions and Use… · p. 18
  • Table 6 reports minimal multi-policy replay results, with GPT-4o limited to a 50-row sample that should not be compared with full-panel models by raw accuracy.Source: Table 6 · p. 23

I Additional Simulation Diagnostics

  • The appendix’s full-process replay of the September 2008 short-selling ban compares diffusion, opinion-dynamics, classifier, and GPT-4o variants across trigger, action, responder, timing, distribution, and cascade-shape measures.Source: I Additional Simulation Diagnostics · p. 23
  • The replay interface is modular, allowing future replacement of decision rules with richer memory, role-conditioned prompts, selective influence, counterfactual interventions, or market feedback.Source: I Additional Simulation Diagnostics · p. 23

I.1 Minimal Multi-Policy Replay

The minimal replay shows that accuracy can conceal missed stakeholder actions: the structured classifier leads on active-F1 and macro-F1, while the GPT-4o agent over-triggers despite higher active recall. Feature-discovery diagnostics identify strong trigger signals and describe the models used to inspect the replay panel.

I.1 Minimal Multi-Policy Replay

  • The bounded GPT-4o replay uses only a 50-row subset, so its raw accuracy is not directly comparable with full-panel models.Source: §I.1 · p. 23
  • The structured classifier has the best active-F1 and macro-F1, while GPT-4o has the highest active recall but substantially over-triggers; no-action baselines’ high accuracy masks near-zero active-action F1.In the post-calibration Policy 5 split, no-action diffusion baselines reach .939 accuracy but zero active-F1; GPT-4o reaches .824 active recall and .122 active-F1, while the structured classifier reaches .069 active-F1 and .176 active recall.Source: §I.1 · pp. 23–24, Figure 14 · p. 23
  • The Markov baseline most closely matches the action distribution, although simple temporal persistence can miss many specific responders.Source: §I.1 · p. 23
  • Identity, prior state, network structure, and neighbor context provide the strongest trigger signal in the diagnostics.Source: Figure 15 · p. 24
  • The feature-discovery study uses balanced logistic regression, L1-regularized logistic regression, and random forests to diagnose predictive PAWS fields rather than introduce another simulator.Source: §I.1 · p. 24

I.2 Feature-Discovery Diagnostics

Feature-discovery diagnostics examine signal for triggering, action-type, and sentiment prediction, while a related-work comparison frames PAWS around jointly preserving policy evidence, stakeholder behavior, entity relevance, and market context.

I.2 Feature-Discovery Diagnostics

  • Action-type diagnostics show identity and article-frequency features dominate, consistent with stable organization-level action-mode tendencies in the current panel.Source: Figure 16 · p. 25
  • Sentiment diagnostics show weak, diffuse signal, consistent with validation findings that policy-event sentiment is often target-dependent and unstable.Source: Figure 17 · p. 25
  • Trigger diagnostics identify identity, prior state, network structure, and neighbor context as the strongest signals.Source: §I.2 · p. 24, Figure 18 · p. 26

J Additional Comparison to Related Work

K Detailed Research Questions and Use Cases

This section describes PAWS’s temporally aligned evidence for studying policy context, stakeholder actions, and market outcomes together, while noting that it does not itself solve causal identification in multi-agent systems. It inventories EventFrame interaction and financial-action categories, their mappings to broader or external taxonomies, and associated attributes across financial, regulatory, monetary-policy, macroprudential, and crisis-response events.

K Detailed Research Questions and Use Cases

  • PAWS supports studying which agents influence others and when, using dated stakeholder and organization actions to analyze multi-turn interactions and selective influence.The section frames this as a research question about explaining agent-to-agent influence, not merely producing aggregate outcomes.Source: K Detailed Research Questions and Use… · p. 28
  • PAWS enables tests of whether simulated outcomes align with agent actions, including how memory, interaction topology, policy conditioning, and temporal update rules affect alignment.Its action types, actor identities, event dates, and surrounding policy context support these tests.Source: K Detailed Research Questions and Use… · p. 28
  • PAWS supports comparing documented policy intentions with stakeholder reactions and market context to study expected and unexpected outcomes, counterfactuals, regime shifts, and second-order effects.The stated research question concerns whether multiagent systems can robustly capture both expected and unexpected policy outcomes.Source: K Detailed Research Questions and Use… · p. 28

L Complete EventFrame Category Inventory

  • The stored value n/a denotes structural ineligibility, not missingness or uncertainty, and is shared across independently gated crosswalks and applicable status or direction fields.CAMEO and ACE child fields inherit n/a from their parent; the inventory omits repetitive rows for this shared value.Source: L Complete EventFrame Category Inventory · p. 29
  • The EventFrame inventory distinguishes communication, intent or plan, requests or pressure, material and institutional actions, and observed market or economic states.The inventory also includes legal or enforcement actions and an “other” fallback when an event lacks sufficient semantic support for a more specific interaction mode.Source: L Complete EventFrame Category Inventory · p. 30
  • The inventory distinguishes public-comment mechanisms such as pessimistic and optimistic comments, warnings, forecasts, denials, disclosures, and testimony.These mechanisms are represented as financial-action subtypes linked to public-statement or related cameo categories.Source: L Complete EventFrame Category Inventory · pp. 31–32
  • Stakeholder response subtypes include proposals, demands, endorsements, criticism, negotiations, agreements, plans, threats, rejection, and protest.Their mappings include appeal or demand, consultation, cooperation intent, disapproval or rejection, protest, and legislative policy authorization categories.Source: L Complete EventFrame Category Inventory · p. 32
  • Federal Reserve policy subtypes cover target-rate changes and rate guidance, plus open-market purchases and sales, repo operations, and reserve-requirement changes.The inventory maps target-rate changes and rate guidance to fed monetary policy rate guidance, and open-market purchases and sales to Fed open-market and reserve operations.Source: L Complete EventFrame Category Inventory · p. 33
  • Liquidity-related subtypes include primary, secondary, and seasonal discount-window credit and central-bank liquidity swaps, mapped to lending or reserve-operation categories.The inventory also lists FIMA repo facilities among open-market and reserve-operation mappings.Source: L Complete EventFrame Category Inventory · p. 33
  • Facility entries map named mechanisms such as term auction and commercial paper funding facilities to Fed lending and liquidity facilities.The inventory also names primary dealer credit, money market liquidity, and other lending facilities.Source: L Complete EventFrame Category Inventory · p. 33
  • Public-support subtypes include bailouts, capital injections, guarantees, asset purchases, tax changes, fiscal spending, and subsidies.Lending support is mapped to both Fed lending and liquidity facilities and public support.Source: L Complete EventFrame Category Inventory · p. 34
  • The inventory also maps regulatory actions such as rule changes and short-selling bans, alongside enforcement investigations, lawsuits, and fines.These entries connect regulatory subtypes to coercive financial regulation and enforcement actions to investigation or legal-enforcement categories.Source: L Complete EventFrame Category Inventory · pp. 34–35
  • The inventory maps settlement and criminal charge to legal-enforcement actions, while bankruptcy filing, protection, liquidation, nationalization, mergers, asset sales, and divestitures map to financial resolution or restructuring.Examples include settlement and criminal charge as investigative or legal enforcement, and bankruptcy filing and liquidation as financial resolution, restructuring, or bankruptcy.Source: L Complete EventFrame Category Inventory · p. 35
  • Market-transaction subtypes include share buybacks, dividend changes, debt and equity issuance, securities purchases and sales, short sales, and loan origination or repayment.The listed subtype mappings place these actions in the broader financial-market transaction category.Source: L Complete EventFrame Category Inventory · pp. 35–36
  • Credit-rating changes, defaults, loss write-downs, and earnings or profit-and-loss changes are grouped as credit-rating or balance-sheet effects.The inventory also assigns liquidity conditions to both market-price or liquidity effects and credit-rating or balance-sheet effects.Source: L Complete EventFrame Category Inventory · p. 36
  • The inventory also distinguishes market-price or liquidity effects, real-activity effects, and legislative-policy authorization, with examples including price moves, employment changes, and law enactment.Legislative votes and policy authorizations are also assigned to legislative-policy authorization.Source: L Complete EventFrame Category Inventory · pp. 36–37
  • CAMEO mappings use the most-specific supported two-to-four-digit official code, falling back to the root when no more detailed hierarchy is available.The inventory describes roots for cooperation, conflict, protest, coercion, and other interaction types.Source: L Complete EventFrame Category Inventory · pp. 37–38
  • EventFrame groups CAMEO-derived interactions into verbal cooperation, material cooperation, verbal conflict, and material conflict, alongside financial-action and market or economic effect categories.The listed financial-action families include monetary policy, liquidity facilities, market transactions, business adjustments, and legislative authorization.Source: L Complete EventFrame Category Inventory · pp. 37–38
  • The Federal Reserve tool inventory covers operations, reserve policies, liquidity facilities, and lending programs, including open-market operations, swap lines, and Main Street Lending.The listed tools include standing repo operations, credit facilities, and programs such as PPPLF and MLF.Source: L Complete EventFrame Category Inventory · pp. 39–40
  • The Federal Reserve tool inventory separately identifies the Term Auction Facility, maturity-extension or reinvestment policy, and other evidenced Fed instruments.The inventory references the Federal Reserve tool catalogue (Board of Governors of the Federal Reserve System 2025).Source: L Complete EventFrame Category Inventory · pp. 40–41
  • The ACE/ERE-style crosswalk covers life, movement, transaction, and business events, including examples such as injury, transport, ownership transfer, and organization formation.It also lists marriage and divorce, money transfers, mergers, bankruptcy, and organization dissolution.Source: L Complete EventFrame Category Inventory · p. 41
  • The crosswalk also covers conflict, contact, personnel, and justice events, including attacks or demonstrations, meetings or broadcasts, position changes, and legal processes.Listed justice examples include arrest, charge, trial, suit, conviction, and fine.Source: L Complete EventFrame Category Inventory · p. 41
  • Central-bank monetary-policy actions are classified as current policy-rate or target-range decisions, communication about the expected future policy path, balance-sheet measures, liquidity operations, and reserve-framework actions.Source: L Complete EventFrame Category Inventory · p. 43
  • The IMF iMaPP macroprudential inventory includes countercyclical and capital-conservation buffers, capital requirements, leverage limits, and loan-loss provisioning.These classes are linked to IMF iMaPP.Source: L Complete EventFrame Category Inventory · p. 43
  • Additional iMaPP classes cover limits on credit growth and loan restrictions conditioned on loan, borrower, or bank characteristics, including foreign-currency lending rules.The inventory cites IMF iMaPP.Source: L Complete EventFrame Category Inventory · p. 44
  • EventFrame maps macroprudential tools such as loan-to-value limits and crisis responses such as deposit freezes and bank nationalizations to IMF iMaPP and crisis-response classifications.The listed crisis-response categories also include restructuring costs, liquidity support, guarantees, asset purchases, and monetary or fiscal expansion.Source: L Complete EventFrame Category Inventory · p. 44
  • Semantic attributes capture whether actions are actual, announced, or planned; lifecycle status; policy or effect direction; and the actor’s institutional role.Roles span public authorities, financial and nonfinancial firms, investors, researchers, media, and other or unknown actors.Source: L Complete EventFrame Category Inventory · pp. 45–47

Related papers

Loading 2609.28547v1…