Live Trading Evidence
Live trading evidence is the performance of a strategy run with real capital in a real market over a disclosed, dated period — as opposed to a backtest, paper-trade or simulation. It is the strongest class of evidence for a trading-strategy profitability claim because it is the only class that cannot be retro-fitted: a live record absorbs real fills, real costs and slippage, real latency, real market impact, regime changes the modeller never saw, and the discipline of not being able to re-run the period after inspecting it. Where a backtest is a claim, a sufficiently long live record (with stated capital, costs, benchmark and drawdowns) is closer to a verified finding.
For this vault the central observation is an absence. Across every Markov model class researched — Markov Chain Trading Model, Hidden Markov Model Regime Detection, Markov Regime-Switching Model, Markov Decision Process Trading Model and Reinforcement Learning Trading Policy — no source provides a credible, disclosed live track record of a standalone Markov-model trading strategy. The literature is dominated by backtests and simulations: Sun Wang An 2021 notes RL trading is trained almost entirely offline because live interaction is risky and impractical, and Buczynski, Cuzzolin and Sahakian’s review of ML investing found the field “conspicuously lacking in high-profile success cases”. The nearest aggregate live data points are negative — the Eurekahedge AI Hedge Fund Index of ML-driven funds underperformed the S&P 500 over fifteen years, and prominent ML funds (Aidyia, Sentient Technologies) liquidated within months.
This absence is not neutral. Because the Backtest-to-Live Performance Gap is large and well-documented, a missing live record is the expected state if the underlying backtests are overfit, cost-blind or selected from an undisclosed search — so “no live evidence” functions as corroborating negative evidence, not merely a data gap. It is also why the vault’s profitability rubric reserves its strong grade for results with independent replication and, ideally, live confirmation: a Markov backtest, however clean, cannot reach the top tier on simulation alone. Quant track records that do exist (e.g. Renaissance Technologies’ Medallion) are not disclosed in a form that attributes returns to a Markov method, so they cannot substantiate a Markov-specific claim either.
Connections
- Backtest-to-Live Performance Gap — lacks_live_evidence, the gap is why missing live evidence is decisive, source: https://pmc.ncbi.nlm.nih.gov/articles/PMC8019690/
- Sim-to-Real Gap — relates, explains why RL Markov models stay in simulation, source: https://arxiv.org/abs/2109.13851
- AI Hedge Fund Index Underperformance — reports_underperformance, the only aggregate live ML-fund data, and it is negative, source: https://www.ig.com/za/prime/insights/articles/has-artificial-intelligences-impact-on-hedge-funds-been-overhype-241121
- Markov Chain Trading Model — lacks_live_evidence, source: https://pmc.ncbi.nlm.nih.gov/articles/PMC8019690/
- Hidden Markov Model Regime Detection — lacks_live_evidence, source: https://pmc.ncbi.nlm.nih.gov/articles/PMC8019690/
- Reinforcement Learning Trading Policy — lacks_live_evidence, source: https://arxiv.org/abs/2109.13851
- Out-of-Sample Backtesting — relates, live evidence is the tier above OOS testing, source: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3104816
Live Trading Evidence [opposes] Backtest-to-Live Performance Gap Sim-to-Real Gap [contradicts] Live Trading Evidence
Sources
- Buczynski, W., Cuzzolin, F., & Sahakian, B. (2021). “A review of machine learning experiments in equity investment decision-making.” International Journal of Data Science and Analytics, 11(3), 221–242. https://pmc.ncbi.nlm.nih.gov/articles/PMC8019690/
- López de Prado, M. (2018). “The 10 Reasons Most Machine Learning Funds Fail.” Journal of Portfolio Management, 44(6), 120–133. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3104816
- IG Prime (2024). “Has artificial intelligence’s impact on hedge funds been overhyped?” https://www.ig.com/za/prime/insights/articles/has-artificial-intelligences-impact-on-hedge-funds-been-overhype-241121