Honest Limitations
Every trading system has boundaries. Understanding what Gordon cannot do is as important as understanding what it can.
What Gordon Is
Gordon is a systematic trend-following system for crypto futures. It:
- Detects trends using mathematical indicators (not predictions)
- Sizes positions based on risk, not conviction
- Manages risk with mandatory stop losses and circuit breakers
- Validates every strategy through walk-forward testing before deployment
- Runs autonomously 24/7 with minimal intervention
What Gordon Is Not
Not a Prediction Engine
Gordon does not predict where prices will go. It reacts to trends that have already started. This means:
- It will always be late to a move (it needs confirmation)
- It will always give back some profit at the end of a move (it needs reversal confirmation)
- It will never catch the exact top or bottom
Not High-Frequency Trading
Gordon operates on timeframes from 1 hour to 1 week. It does not:
- Trade on sub-minute data
- Use order book depth or tick data (except for VPIN)
- Compete on latency
- Execute hundreds of trades per day
Not a Guaranteed Profit Machine
Historical backtests show positive expected returns, but:
- Past performance does not guarantee future results
- The crypto market is changing (ETF institutionalization compresses volatility)
- Black swan events can exceed historical drawdown levels
- The system can and will have losing months
Not a Black Box
All validated strategies are interpretable:
- Supertrend: "price crossed above/below an ATR band"
- EWMAC: "fast moving average crossed slow moving average"
- PSAR: "parabolic dots flipped to other side of price"
There is no deep learning, no neural network, no unexplainable signal.
Known Limitations
Simulation vs Reality
Backtests are simplified models of reality. Gordon's backtests include fees and slippage estimates, but cannot fully account for:
- Market impact — large orders move the price (less relevant for $5-10k accounts)
- Exact fill prices — backtests assume fills at the close, reality is messier
- Exchange downtime — Binance has occasional outages
- API rate limits — may delay order execution
Data Limitations
- All data comes from Binance — no multi-exchange coverage
- Spot data used for maximum history, but futures behavior may differ slightly
- 1-minute candles are the finest granularity — no tick data for most analysis
Structural Changes
The crypto market is structurally different from when many of our backtests begin:
- 2017-2019: retail-dominated, high volatility, easy trends
- 2020-2023: DeFi, institutional entry, still volatile
- 2024+: ETF era, reduced volatility, more efficient
Strategies that worked well in 2017 may work less well going forward. Gordon addresses this through multi-strategy diversification and ongoing research (v3 alpha-seeking).
Single Exchange Risk
Gordon runs on Binance. If Binance experiences:
- Extended downtime
- Regulatory shutdown
- Security breach
The system cannot trade. Mitigation: regular withdrawals to reduce exchange exposure.
The Honest Expected Outcome
Based on walk-forward testing across 5-8 years of data:
| Metric | Realistic expectation |
|---|---|
| Annual return | 15-25% in good years, -10-15% in bad years |
| Max drawdown | Up to 25% from peak |
| Win rate | 40-50% (profitable through R:R > 2:1) |
| Losing streaks | 5-8 consecutive losses is normal |
| Time underwater | Months at a time during sideways markets |
This is not exciting. It is designed to be boring and reliable over years, not months.