cs.AISep 4, 2026

Artificial Intelligence in Equity and Crypto Markets: Progress, Profitability Evidence, and the Limits of Automated Investing

Authors: Linsen Zhu, Mengqing Cai

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Artificial intelligence (AI) now supports investment workflows from data and prediction through research, portfolios, execution, and tool use. Technical capability, however, is not evidence of investment profitability. This critical state-of-the-art review examines public research available through 31 August 2026 on listed equities, exchange-traded funds, centralized crypto spot, perpetual futures, and on-chain markets. We organize evidence with an alpha-translation chain: point-in-time information must yield a stable signal, feasible positions, executable orders, and risk-adjusted returns after costs. Across machine learning, time-series foundation models, financial language models, reinforcement learning, and agents, the examined record shows real but mainly upstream progress in prediction, text processing, portfolio design, and workflow integration. Evidence is thinner for durable net performance. Temporal contamination, repeated selection, survivorship, weak benchmarks, implementation costs, venue mechanics, and capacity can break translation to net alpha. Strong historical results coexist with predictor decay, corrected look-ahead failures, mixed prospective evidence, and few audited live-capital records. Crypto adds informative state but requires separate treatment of spot, perpetual, and decentralized cash flows and execution. Within the public evidence examined here, no general AI architecture is shown to deliver persistent, cross-regime, capacity-aware net alpha. More credible claims require point-in-time data and models, decision-aligned objectives, joint portfolio--execution evaluation, controlled adaptation, prospective tests, and authority-matched governance. These conditions can improve evidence and implementation; they do not guarantee profit.

Figures & tables

Figure 1: The alpha-translation chain. Technical improvement at one stage is economically valuable only to the extent that information survives all downstream transformations. Red labels identify representative failure channels, not an exhaustive list.
CodeEvidence objectPrimary supportDimensions unresolved unless reported
CAPArchitecture or vendor capabilityFeasibility of representation, tool use or order connectivityPrediction, portfolio mapping and return
TASKTask metricClassification, extraction, reasoning or forecast accuracy on a defined benchmarkFeasible positions, implementation and economic value
OOSTime-valid statistical testIncremental prediction after a declared historical splitPortfolio value, costs and capacity
GROSSGross portfolio simulationHistorical decision value under stated weights and constraintsImplementable net performance
NETNet, risk-adjusted simulationSimulated alpha under disclosed cost, financing and execution modelsProspective robustness and unmodeled costs
PROSPPrecommitted paper or shadow portfolioResistance to hindsight and changing model selectionReal fills, operational loss and live-capital behavior
Table 1: Evaluation-design archetypes within a multidimensional profitability-evidence profile. Rows are common study designs, not a total ordering; each must be assessed on all seven dimensions in Equation 2 .
FamilyDemonstrated progressPrincipal economic opportunityCharacteristic evidence gap
Regularized ML, trees, boostingNonlinear and high-dimensional return prediction in large panelsRanking and interaction discoveryMultiple testing, unstable features, liquidity concentration
Deep asset-pricing modelsJoint representation and economic restrictionsLatent states, nonlinear stochastic discount factorsEffective sample size, interpretation, historical dependence
TSFMsZero-shot and cross-domain forecast transferFast adaptation, risk and covariance inputsGeneric loss is not alpha; pretraining contamination
Financial NLP/LLMsScalable extraction and semantic event processingTimely news, filing and expectation signalsMemorization, timestamping, prompt selection, source provenance
RL and direct policiesSequential or decision-aligned objectivesTurnover, constraints and dynamic risk inside learningSimulator error, reward hacking, fragile policy transfer
LLM agentsIntegration of retrieval, memory, reasoning and toolsEnd-to-end research and operational coordinationComponent attribution, reproducibility, cost and live validation
Table 2: Method families viewed through the alpha-translation chain. “Progress” denotes an evidence-backed technical contribution, not a general profitability finding.
System/sourceMarket emphasisIntegration contributionProfitability boundary
FinMem ( Yu et al., 2023 )EquitiesLayered memory and reflective trading workflowRetrospective experiments; memory benefit is conditional on protocol
FinAgent ( Zhang et al., 2024 )EquitiesMultimodal observations, tools and generalist agentHistorical evaluation, not independent live-capital persistence
FinCon ( Yu et al., 2024 )Equities/other tasksMulti-agent roles and conceptual verbal reinforcementReported backtests do not isolate a general alpha mechanism
TradingAgents ( Xiao et al., 2024 )Selected US equitiesAnalyst debate, trader and risk/portfolio rolesShort historical test on selected stocks; architecture is not durable-alpha proof
StockBench ( Chen et al., 2025 )EquitiesTime-structured benchmark for LLM-agent tradingBenchmark outcome remains model-, window- and cost-specific
Agent Trading/Market Arenas ( Ma et al., 2025 ; Qian et al., 2026 )Controlled and multi-marketComparative numerical behavior and live-market evaluationProspective arenas improve validity but need longer, fill-aware, risk-adjusted records
Table 3: Representative agentic integration systems and interfaces. The table describes the principal evidenced contribution, not a performance ranking.
StratumDistinct information/data riskDistinct implementation lossMinimum economic validation
Listed equities/ETFsPublication lags, revisions, constituent and delisting bias, corporate actionsSpread, impact, commission, borrow, auction/halts; ETF premium and path dependenceLiquidity/capacity slices, borrow and turnover, factor alpha, corporate-action-correct universe
Centralized crypto spotChanging token/venue universe, wash volume, forks, quote-asset and API revisionsFragmented fills, custody, prefunding, withdrawal and stablecoin riskVenue-level executable prices, survival-free universe, custody/capital charge, manipulation sensitivity
Perpetual futuresHistorical contract specifications, index/mark construction, changing funding rulesFunding, leverage, margin, liquidation, insurance/auto-deleveragingContract-level funding and mark-price path, collateral and margin rules, tail liquidation tests
On-chain/DEXState at signing versus inclusion, address labels, reorg/finality, adversarial activityGas/priority fees, slippage, failed calls, MEV, LVR, smart-contract lossOrdered transaction simulation, all failed/successful gas, rebalancing benchmark, contract/allowance controls
Table 4: Market strata require different P&L and validity fields. Items in the final column are minimum additions to ordinary return reporting.
EvidenceEvaluation modeAffirmative inferenceBinding qualification
Gu et al.; Deep SDFHistorical equity OOSNonlinear characteristic structure improves selected forecasts/portfoliosHistorical, universe- and implementation-specific
Deep policies; AlphaPortfolioHistorical decision OOSEconomic objectives and sequence modeling can improve selected portfoliosPreference, search, constraint and cost-model dependence
Implementable frontierHistorical cost-aware OOSJoint cost-aware learning dominates cost-agnostic construction in the studySimulated costs and capacity; not live capital
Post-cutoff news LLMPost-training-cutoff event testLLM semantics can identify return-relevant news and later driftLatency, adoption decay and portfolio implementation
Point-in-time LMsChronological model checkpointsUseful text signals need not rely on future-trained modelsHistorical portfolio construction remains model-specific
Prospective household portfoliosDaily precommitted recommendationsDirect evidence of LLM style, concentration and behaviorNo significant adjusted abnormal return; preliminary period
Table 5: Representative evidence and the maximum inference it supports. Numerical results are not compared across rows because universes, horizons, objectives and costs differ.
More in this topic

Algorithmic Trading

Algorithmic trading uses computational models to generate signals, allocate portfolios, and execute orders across financial markets.

See all 149 papers

Explore similar work

CardsList
Sep 28, 2026

Can AI Make Money in Crypto? Measuring the Gap from Backtests to Real Markets

AI-based trading methods have rapidly evolved from machine learning and reinforcement learning to large language models (LLMs) and trading agents, yet their performance is still predominantly assessed through historical backtesting. Such evaluations provide limited evidence of whether a method can generalize to unseen future markets or whether its backtested performance can be sustained in realistic trading frictions (e.g., latency, slippage, liquidity constraints, and market impact). We present a unified benchmark that evaluates representative machine learning, reinforcement learning, LLM-based, and agent-based trading methods in cryptocurrency markets through three progressively more realistic stages: historical backtesting, prospective exchange-based paper trading, and real-money live trading. These stages jointly increase temporal realism by moving from historical to unseen future markets, and execution realism by moving from offline simulation toward live trading. This protocol enables us to quantify the backtest-to-realization gap, identify when performance begins to deteriorate, and compare how this gap differs across major classes of AI trading methods. We further provide a unified open-source system supporting all three evaluation stages, together with a public platform that continuously updates benchmark results. Code is available at https://github.com/Starlien95/Awesome-TradingAI.

Xingtong Yu, Jiarun Zhou, Guanlin Ding +12

The Chinese University of Hong Kong · University of Science and Technology of China · Jilin University +3

Algorithmic Trading
AskPDF
Jul 21, 2026

Predictive Extrema, Unprofitable Policies: An AI-Assisted Audit of Candle-Based Binance Spot Timing Models

We audit whether candle-based machine-learning models can turn predictions of cryptocurrency extrema or short-horizon outcomes into positive Binance Spot paper policies after assumed costs. Numerical results come from scripted fixed-seed model runs and deterministic simulators; human-supervised AI agents supported the July 20 evidence-integrity revision through literature retrieval, separately tasked critique, artifact reconciliation, documentation, and source packaging, not trading decisions. The strongest later-period evidence, conditional on extensive predecessor search, is negative: an unchanged ten-pair mandatory-daily selector lost 6.72% over 19 July cycles at an assumed 31-bps completed-cycle cost, with 3 wins and 16 losses. In short model-specific July evaluations, the validation-selected local-minimum policy returned -1.79%, while the local-maximum sell-to-cash/re-entry policy underperformed continuous holding by 2.80%; their gross mean advantages of 11.11 and 12.21 bps were below even the 21-bps stress. A Gurgul-inspired, OHLCV-only daily adaptation attained minimum/maximum ROC AUC of 0.874/0.896 but average precision of only 0.134/0.116 and lost 44.30% over seven cycles, versus -41.20% for buy-and-hold. A forensic audit also downgraded an earlier One4All "30-day holdout": its dates had influenced prior architecture work, its four-hour outcome horizon was not purged at split boundaries, it used same-close entry, and its raw result directories were absent. Across the tested, mostly exploratory protocols, event-ranking performance did not establish positive executable policy value. Every operational decision remains NO_TRADE.

Ayoub Jadouli

Computer Science and Smart Systems, Faculty of Sciences and Technology Abdelmalek Essaâdi University, Tangier, Morocco

Quantitative FinanceAlgorithmic AuditingFinancial Time SeriesFinanceAlgorithmic Trading
AskPDF
Jun 11, 2026

Crypto x AI, AI x Crypto: A Survey

The intersection of crypto x AI is spawning papers, products, online posts, and companies. All the surrounding buzz, though, obscures what exactly has been done, what the opportunities and challenges are, and what open questions deserve attention. This survey paper asks what AI can do for blockchain-based technologies (broadly construed as "crypto") (crypto x AI), and vice versa (AI x crypto). We systematize existing work, summarize key takeaways, highlight open research questions, and offer a perspective on pervasive industry misconceptions, concluding that AI and crypto are still in the very early stages of meaningful integration.

Sarah Allen, Pranay Anchuri, James Austgen +22

1Initiative for CryptoCurrencies and Contracts (IC3) · 5Flashbots · 6Offchain Labs +11

Topic pending
AskPDF