For three decades, systematic quantitative strategies have been the preserve of institutional trading desks. The infrastructure required to run them — real-time data pipelines, regime-aware allocation, ML signal filtering — has been prohibitively complex and expensive to build. FerroQuant removes that barrier entirely.
The professional trading industry has known for decades that systematic, rule-based approaches outperform discretionary decision-making over any sufficiently long time horizon. The evidence is not theoretical — it is embedded in the returns of the world's most successful trading firms, from Renaissance Technologies to Winton Group to Two Sigma.
The reason is straightforward. Human cognition is poorly suited to consistent execution under uncertainty. Loss aversion causes premature exits. Recency bias distorts probability estimates. Overconfidence inflates position sizes after winning streaks. These effects compound. Barber and Odean (2000) documented that the most active retail traders earn the lowest returns — not from lack of intelligence, but from the irreducible costs of discretionary decision-making at scale. [Barber, B.M. & Odean, T. (2000). Trading Is Hazardous to Your Wealth. Journal of Finance, 55(2), 773–806.]
The second problem is scope. No human trader can meaningfully monitor over a thousand instruments across five asset classes simultaneously. By the time a manual analysis is complete, the entry opportunity has passed or the conditions have changed. Systematic execution at the speed of computation is not a marginal advantage — it is a categorical one.
The third problem is market regime. Most strategies, however well-designed, are regime-specific. A momentum strategy that compounds effectively during a trending market will suffer significant drawdown in a mean-reverting environment. Without a mechanism to detect regime transitions and adapt strategy allocation accordingly, even sophisticated approaches degrade over time.
The institutions solved these problems long ago. The infrastructure they built to do so has, until now, been inaccessible to individual traders. FerroQuant is built on the same principles.
Quantitative finance is not an esoteric discipline. It is a body of peer-reviewed research spanning four decades, accumulated across some of the most rigorous academic institutions in the world. The following principles form the foundation of FerroQuant's methodology.
Momentum. Jegadeesh and Titman (1993) demonstrated that securities which have outperformed over a 3–12 month window continue to outperform in the near term — a finding subsequently replicated across equities, bonds, currencies, and commodities. Moskowitz, Ooi and Pedersen (2012) extended this to time-series momentum, showing that the asset's own past return predicts its future return with statistical significance across 58 liquid instruments. [Jegadeesh, N. & Titman, S. (1993). Returns to Buying Winners and Selling Losers. Journal of Finance, 48(1), 65–91. | Moskowitz, T., Ooi, Y.H. & Pedersen, L.H. (2012). Time Series Momentum. Journal of Financial Economics, 104(2), 228–250.]
Mean reversion. De Bondt and Thaler (1985) showed that extreme price movements tend to reverse over longer horizons. At shorter intraday and daily timeframes, the same principle holds in range-bound markets where volatility is compressed. The critical insight is that momentum and mean reversion are not competing hypotheses — they operate in different market states, and the key determinant is regime. [De Bondt, W.F.M. & Thaler, R. (1985). Does the Stock Market Overreact? Journal of Finance, 40(3), 793–805.]
Ensemble diversification. Markowitz (1952) established that combining uncorrelated return streams reduces portfolio variance without proportionally reducing expected return. When applied to strategy portfolios rather than asset portfolios, the mathematics is identical: running 178 strategies with differentiated edge profiles materially improves signal-to-noise ratio relative to any single strategy in isolation. [Markowitz, H. (1952). Portfolio Selection. Journal of Finance, 7(1), 77–91.]
Regime detection. Hamilton (1989) demonstrated that macroeconomic time series exhibit regime-switching behaviour — periods of distinct statistical properties separated by transition points that can be modelled using Hidden Markov Models. Applied to financial markets, this framework allows systematic identification of trending, ranging, and volatile regimes, enabling appropriate strategy allocation for each condition. [Hamilton, J.D. (1989). A New Approach to the Economic Analysis of Nonstationary Time Series. Econometrica, 57(2), 357–384.]
Each of these findings has been replicated across asset classes, time periods, and geographies. The evidence for systematic, diversified, regime-aware trading is not speculative — it is among the most robust empirical findings in the financial economics literature.
FerroQuant operationalises the above principles in a production system running continuously across five asset classes. Each layer of the architecture corresponds directly to an empirically validated concept.
Continuous Hidden Markov Model analysis identifies the prevailing market state across all five asset classes. Strategy allocation is conditioned on regime classification. Strategies incongruent with the current regime are suppressed before signal evaluation.
178 quantitative strategies compete continuously on live market data using successive halving elimination. Strategies are evaluated on risk-adjusted performance metrics. Only those demonstrating statistically significant edge under current conditions progress to signal generation.
Candidate signals pass through an ONNX ensemble model trained separately for long and short directional bias. The filter is calibrated to minimise false positives rather than maximise recall — fewer, higher-conviction signals are preferable to a high-volume, low-quality signal stream.
Six independent signal engines — one per market — run in parallel. Cross-asset correlations are computed in real-time. Signal confidence is adjusted for correlated exposures, preventing inadvertent concentration risk across notionally distinct instruments.
The following table presents a factual comparison of FerroQuant's current production capabilities against the leading alternatives available to retail and institutional participants.
| Capability | FerroQuant | Numerai | QuantConnect | Self-built |
|---|---|---|---|---|
| Asset class coverage | 5 classes | Equities only | 9 classes (cloud) | Dependent on build |
| Live HMM regime detection | Production | None | Not provided | Must be built |
| Strategy tournament system | Live, 178 strategies | Encrypted submissions | Not provided | Must be built |
| ML signal filtering | ONNX ensemble | Yes | Limited | Must be built |
| Real market data access | Full, unencrypted | Obfuscated | Yes | Yes |
| Infrastructure overhead for users | None | None | Moderate | 2–3 years to build |
| Live execution | 24/7 production | Payouts only | Yes | Yes (if built) |
FerroQuant's infrastructure operates at a scale that surfaces empirical questions not easily studied in academic or synthetic settings. The following represent active areas of investigation within the platform. We welcome collaboration with quantitative researchers who have relevant expertise.
Researchers interested in accessing FerroQuant's production infrastructure and historical dataset for collaborative investigation are invited to register on the platform and indicate their institutional affiliation and research focus.
The methodology outlined above runs continuously in production. Free access is available. Professional and Trader tiers unlock the full signal stream, priority execution data, and advanced analytics.
This document is provided for informational purposes only and does not constitute financial advice or a solicitation to trade. Past performance of systematic strategies does not guarantee future results. All trading involves risk of loss.