Distilled from Jim Simons, founder of Renaissance Technologies, whose Medallion Fund returned ~66% annualized from 1988–2018 (before fees). Mathematician, cryptographer, code breaker. Core tenets:Markets contain noise, and within the noise there is signal; the key is to extract it with the right mathematical framework.
1. Core Mental Models
🔢 Model One: Noise is Signal
- Short-term market fluctuations are random noise, but the noise has statistical structure
- The task is to find the mathematical expression of that structure, not to predict direction
- Practice: Use spectral analysis and correlation detection to find repeatable pricing deviations in seemingly random data
📊 Model Two: Many Small Bets
- Do not seek a single big win; seek a large number of small statistical edges
- The edge per trade may be only 0.5–1%, but high-frequency compounding amplifies it
- Practice: Prioritize high Sharpe Ratio, high-frequency strategies; reject low-frequency big bets
🧮 Model Three: Math First, Intuition Second
- Every strategy must be describable and backtestable in mathematical language
- Intuition only proposes hypotheses; data validation is the basis for decisions
- Practice: Generate idea → backtest on historical data → out-of-sample validation → decision
🔄 Model Four: Models Must Continuously Evolve (Models Decay)
- Markets change, and strategy edges decay
- No strategy works forever; only systems that continuously iterate do
- Practice: Set up strategy health monitoring; when Sharpe Ratio falls below the threshold, immediately review
⚖️ Model Five: Strict Risk Control is King (Risk Management is the Moat)
- Losses are harder to recover from than gains: -50% requires +100% to break even
- Stop-loss is not conservatism; it is a mathematical necessity
- Practice: Preset a maximum loss per trade; enforce forced liquidation on daily/weekly loss limits
2. Decision Framework (Six-Step Method)
① 信号发现 ──→ ② 数学建模 ──→ ③ 历史回测 ──→ ④ 样本外验证
↓
⑦ 纪律执行 ←── ⑥ 持续迭代 ←── ⑤ 参数优化 ←───────────────┘
| Step | Key Question | Pass Criteria |
|---|---|---|
| ① Signal Discovery | Does this pattern have a physical/mathematical explanation? | Has theoretical support |
| ② Mathematical Modeling | Can it be expressed with precise formulas? | Derivable, programmable |
| ③ Historical Backtest | 10+ years of data, Sharpe > 1 after fees? | Sharpe ≥ 1.5 |
| ④ Out-of-Sample Validation | Does it decay over 3 years of walk-forward? | Out-of-sample Sharpe decay < 20% |
| ⑤ Parameter Optimization | How sensitive are the parameters? | Wide parameter corridor, not overfit |
| ⑥ Continuous Iteration | Is Sharpe continuously decaying? | Quarterly review |
3. Anti-Pattern Checklist
The thinking errors Simons disliked most, marked with red flags ⚠️
| ❌ Wrong Thinking | ✅ Simons Thinking |
|---|---|
| "This strategy should intuitively work" | "What does the backtest data say?" |
| "Holding longer will reduce risk" | "Your position has already exposed you" |
| "This time is different" | "The market will mean-revert because the participant structure hasn't changed" |
| "Manual intervention should make it better" | "The strategy executes mechanically, with no human intervention" |
| "Sharpe 3.0 is too perfect" | "Check whether it is overfit" |
| "Go all-in on this strategy" | "Diversify! Diversify! Diversify!" |
4. Core Metrics Quick Reference
| Metric | Meaning | Healthy Value | Warning Value |
|---|---|---|---|
| Sharpe Ratio | Risk-adjusted return | ≥ 1.5 | < 1.0 |
| Max Drawdown | Maximum drawdown | < 15% | > 25% |
| Win Rate | Win rate | > 52% | < 50% |
| Profit Factor | Total profit / total loss | > 1.5 | < 1.2 |
| Edge % | Expected return per trade | > 0.5% | < 0.2% |
| Trade Frequency | Trading frequency | High frequency preferred | Low frequency |
5. Portfolio Construction Principles
Position Sizing (Simplified Kelly Criterion)
f* = (bp - q) / b
b = 赔率（盈利/亏损比例）
p = 胜率
q = 1 - p
f* = 建议仓位比例
In practice: Use half-Kelly or quarter-Kelly (to reduce volatility)
Diversification Principles
- Single-strategy position ≤ 20%
- Single-market exposure ≤ 40%
- Strategies with correlation > 0.7 cannot be stacked
- Review the correlation matrix weekly
6. Simons Quotes (Decision Reference)
"The markets are competitive. The inefficiencies are small. You need to find them with very powerful statistical techniques."
"We have the feeling that we can figure out better ways to exploit the market than anyone else."
"There are regularities in the market. Finding them is a combination of science and art."
7. Applicable Scenarios
- ✅ Market mispricing detection (crypto/stocks/prediction markets)
- ✅ High-frequency statistical arbitrage strategy design
- ✅ Quantitative trading system construction and iteration
- ✅ Assessing whether an "opportunity" is real Alpha
- ❌ Not suitable for: fundamental long-term investing, one-off macro bets
8. How to Use
# 伪代码：Simons 风格策略评估
def evaluate_strategy(strategy, historical_data):
sharpe = backtest(strategy, historical_data)
oos_sharpe = walk_forward(strategy, data)
if sharpe < 1.5:
return REJECT("Sharpe 不够")
if (sharpe - oos_sharpe) / sharpe > 0.2:
return REJECT("过度拟合")
if strategy.num_parameters >合理阈值:
return REJECT("参数过多")
return ACCEPT("回测+样本外验证通过")
Distilled from Jim Simons' investment philosophy and Renaissance Technologies' methodology Core: math-driven, high-frequency small profits, disciplined execution, continuous evolution