Artificial Intelligence in Equity and Crypto Markets: Progress, Profitability Evidence, and the Limits of Automated Investing
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
| Code | Evidence object | Primary support | Dimensions unresolved unless reported |
|---|---|---|---|
| CAP | Architecture or vendor capability | Feasibility of representation, tool use or order connectivity | Prediction, portfolio mapping and return |
| TASK | Task metric | Classification, extraction, reasoning or forecast accuracy on a defined benchmark | Feasible positions, implementation and economic value |
| OOS | Time-valid statistical test | Incremental prediction after a declared historical split | Portfolio value, costs and capacity |
| GROSS | Gross portfolio simulation | Historical decision value under stated weights and constraints | Implementable net performance |
| NET | Net, risk-adjusted simulation | Simulated alpha under disclosed cost, financing and execution models | Prospective robustness and unmodeled costs |
| PROSP | Precommitted paper or shadow portfolio | Resistance to hindsight and changing model selection | Real fills, operational loss and live-capital behavior |
| Family | Demonstrated progress | Principal economic opportunity | Characteristic evidence gap |
|---|---|---|---|
| Regularized ML, trees, boosting | Nonlinear and high-dimensional return prediction in large panels | Ranking and interaction discovery | Multiple testing, unstable features, liquidity concentration |
| Deep asset-pricing models | Joint representation and economic restrictions | Latent states, nonlinear stochastic discount factors | Effective sample size, interpretation, historical dependence |
| TSFMs | Zero-shot and cross-domain forecast transfer | Fast adaptation, risk and covariance inputs | Generic loss is not alpha; pretraining contamination |
| Financial NLP/LLMs | Scalable extraction and semantic event processing | Timely news, filing and expectation signals | Memorization, timestamping, prompt selection, source provenance |
| RL and direct policies | Sequential or decision-aligned objectives | Turnover, constraints and dynamic risk inside learning | Simulator error, reward hacking, fragile policy transfer |
| LLM agents | Integration of retrieval, memory, reasoning and tools | End-to-end research and operational coordination | Component attribution, reproducibility, cost and live validation |
| System/source | Market emphasis | Integration contribution | Profitability boundary |
|---|---|---|---|
| FinMem ( Yu et al., 2023 ) | Equities | Layered memory and reflective trading workflow | Retrospective experiments; memory benefit is conditional on protocol |
| FinAgent ( Zhang et al., 2024 ) | Equities | Multimodal observations, tools and generalist agent | Historical evaluation, not independent live-capital persistence |
| FinCon ( Yu et al., 2024 ) | Equities/other tasks | Multi-agent roles and conceptual verbal reinforcement | Reported backtests do not isolate a general alpha mechanism |
| TradingAgents ( Xiao et al., 2024 ) | Selected US equities | Analyst debate, trader and risk/portfolio roles | Short historical test on selected stocks; architecture is not durable-alpha proof |
| StockBench ( Chen et al., 2025 ) | Equities | Time-structured benchmark for LLM-agent trading | Benchmark outcome remains model-, window- and cost-specific |
| Agent Trading/Market Arenas ( Ma et al., 2025 ; Qian et al., 2026 ) | Controlled and multi-market | Comparative numerical behavior and live-market evaluation | Prospective arenas improve validity but need longer, fill-aware, risk-adjusted records |
| Stratum | Distinct information/data risk | Distinct implementation loss | Minimum economic validation |
|---|---|---|---|
| Listed equities/ETFs | Publication lags, revisions, constituent and delisting bias, corporate actions | Spread, impact, commission, borrow, auction/halts; ETF premium and path dependence | Liquidity/capacity slices, borrow and turnover, factor alpha, corporate-action-correct universe |
| Centralized crypto spot | Changing token/venue universe, wash volume, forks, quote-asset and API revisions | Fragmented fills, custody, prefunding, withdrawal and stablecoin risk | Venue-level executable prices, survival-free universe, custody/capital charge, manipulation sensitivity |
| Perpetual futures | Historical contract specifications, index/mark construction, changing funding rules | Funding, leverage, margin, liquidation, insurance/auto-deleveraging | Contract-level funding and mark-price path, collateral and margin rules, tail liquidation tests |
| On-chain/DEX | State at signing versus inclusion, address labels, reorg/finality, adversarial activity | Gas/priority fees, slippage, failed calls, MEV, LVR, smart-contract loss | Ordered transaction simulation, all failed/successful gas, rebalancing benchmark, contract/allowance controls |
| Evidence | Evaluation mode | Affirmative inference | Binding qualification |
|---|---|---|---|
| Gu et al.; Deep SDF | Historical equity OOS | Nonlinear characteristic structure improves selected forecasts/portfolios | Historical, universe- and implementation-specific |
| Deep policies; AlphaPortfolio | Historical decision OOS | Economic objectives and sequence modeling can improve selected portfolios | Preference, search, constraint and cost-model dependence |
| Implementable frontier | Historical cost-aware OOS | Joint cost-aware learning dominates cost-agnostic construction in the study | Simulated costs and capacity; not live capital |
| Post-cutoff news LLM | Post-training-cutoff event test | LLM semantics can identify return-relevant news and later drift | Latency, adoption decay and portfolio implementation |
| Point-in-time LMs | Chronological model checkpoints | Useful text signals need not rely on future-trained models | Historical portfolio construction remains model-specific |
| Prospective household portfolios | Daily precommitted recommendations | Direct evidence of LLM style, concentration and behavior | No significant adjusted abnormal return; preliminary period |
Algorithmic Trading
Algorithmic trading uses computational models to generate signals, allocate portfolios, and execute orders across financial markets.
See all 149 papers