The wrong strategy in the wrong regime loses money. GateKeeper classifies markets into four states and assigns the right strategy to each — backed by decades of academic research.
Markets cycle between distinct regimes: trending up, trending down, ranging, and crisis. Each regime favours a different type of strategy. Trend-following thrives in trends but gets whipsawed in ranges. Mean reversion thrives in ranges but fails catastrophically in crises. Pairs trading breaks down when correlations spike.
GateKeeper is our regime classifier. It runs every 4 hours, reads market structure across multiple reference instruments, and tells each engine what it’s allowed to do. The goal isn’t to predict the future — it’s to avoid running the wrong strategy in the wrong environment.
Market is in a sustained uptrend. Trend-following strategies (Donchian breakout) are enabled. Mean reversion is disabled — buying dips in a strong trend works until it doesn’t, and the downside when it doesn’t is severe. Research: Hurst, Ooi & Pedersen (2017) found trend-following delivers positive returns across 137 years of data, with slightly higher performance during bull markets.
Market is in a sustained downtrend. Short-side trend-following is enabled. Moskowitz, Ooi & Pedersen (2012) documented significant returns from time-series momentum on the short side. However, bear markets are roughly 2× more volatile than bull markets (Quantified Strategies, 2025), so position sizing accounts for the elevated risk.
Market is oscillating without directional conviction (ADX < 20). Trend-following is disabled — this is one of the most robust findings in trading research. Breakout strategies generate whipsaw losses in range-bound markets (Wilder, 1978). Mean reversion and pairs trading are enabled, as prices bounce reliably between support and resistance.
VIX exceeds 30. Correlations spike, liquidity withdraws, forced selling dominates. Mean reversion fails because the mean itself shifts (QuestDB, 2025). Pairs trading breaks down as correlations that the strategy depends on become unstable (BIS, 2000). Only high-conviction setups are permitted. Position sizes are reduced.
The VIX (CBOE Volatility Index) is the most widely used market fear gauge. The Federal Reserve Bank of St. Louis describes it as a barometer of market uncertainty, with a long-term average around 19–20 and spikes during crisis episodes.
The VIX > 30 threshold for CRISIS detection is well-established in the literature. This level has been reached during the 2008 Global Financial Crisis, the March 2020 COVID crash, and the August 2024 carry trade unwind (BIS Bulletin No. 95, 2024). VIX exhibits strong mean reversion, supporting the use of cooldown periods before returning to normal operations.
Key finding: Harvey et al. (2018) at Man Group showed that scaling positions inversely with volatility improves Sharpe ratios and reduces left-tail events across 60+ assets. This research won the Bernstein Fabozzi Outstanding Article Award.
Our regime design is grounded in nine key findings from the academic and practitioner literature:
1. Trend-following works in trending markets and suffers in ranges (whipsaw).
2. Mean reversion works in ranges and fails catastrophically in crisis.
3. Pairs trading breaks down when correlations spike during crisis.
4. Position sizing should scale inversely with volatility.
5. VIX is a reliable regime proxy for strategy selection.
6. Regime evaluation frequency should match the trading timeframe.
7. Neither trend-following nor mean reversion produces robust returns on crypto perps at 1H–daily timeframes.
8. Momentum works on large-cap crypto; reversal on small-cap.
9. RSI works as momentum confirmation on crypto, not as mean reversion.
Each engine uses a different reference instrument for regime classification, matched to the market it trades:
Engine 1 (Crypto): BTC composite (BTC + ETH + SOL majority vote)
Engine 2 (Futures): MES (Micro E-mini S&P 500) via IBKR
Engine 3 (Forex): DXY (US Dollar Index) via yfinance
This ensures each engine responds to the regime of its own market, not a global proxy that might not be relevant. Crypto and equities can be in different regimes simultaneously.
Through extensive backtesting and live validation, we’ve identified exactly one regime adjustment that consistently adds value: blocking Donchian breakout trades in RANGE regime for Engine 2. This single rule saves approximately 20R of drawdown annually by preventing whipsaw losses.
Other adjustments we tested and removed:
Risk scaling: Reducing position sizes based on regime always created drag.
Removed permanently.
CRISIS blocking for trend strategies: Counter-intuitively, CRISIS trades are
the most profitable (76% win rate for Donchian, 87.5% for MR). Blocking them
cost us money. Removed permanently.
Hurst, Ooi & Pedersen (2017). “A Century of Evidence on Trend-Following Investing.” AQR Capital Management.
Moskowitz, Ooi & Pedersen (2012). “Time Series Momentum.” Journal of Financial Economics, 104(2), 228–250.
Wilder, J.W. (1978). New Concepts in Technical Trading Systems. Trend Research.
Brunnermeier, Nagel & Pedersen (2008). “Carry Trades and Currency Crashes.” NBER Macroeconomics Annual, 23, 313–348.
Harvey et al. (2018). “The Impact of Volatility Targeting.” Journal of Portfolio Management, 45(1), 14–33.
Engle & Granger (1987). “Co-Integration and Error Correction.” Econometrica, 55(2), 251–276.
BIS Bulletin No. 95 (2024). “Anatomy of the VIX Spike in August 2024.”
BIS Conference Paper (2000). “Evaluating Correlation Breakdowns During Periods of Market Volatility.”
Cong, Li & Wang (2021). “Tokenomics: Dynamic Adoption and Valuation.” Review of Financial Studies, 34(3).
The complete research document with 47 citations and detailed per-section analysis is maintained internally and updated as new evidence emerges.