⚠️ Personal research and trading journal — not investment advice. The author does not provide licensed advisory services.
My main trading system involves charts. Pattern recognition, base structure, pivot points, volume signature at breakout. The human read is central.
So I built a parallel system that ignores all of that.
The Simons Branch
The Simons branch — named after Jim Simons, who famously refused to use any discretionary judgment in his Medallion Fund — is a pure quantitative ensemble. No charts. No human reads. Just factors, signals, and rules.
The design: - Universe: US stocks + Thai SET50/100, updated monthly - Signal ensemble: 5 quantitative signals combined with equal weight - Rebalance frequency: 5-day (weekly) - Risk management: same kill-switch and position-sizing rules as the pattern branch
The five signals combine momentum (RS rating, price-to-52w-high), quality (earnings growth slope), and structural factors (volume trend, cross-sectional momentum vs sector). Each signal is z-scored against the universe before combination. Equal weighting rather than optimized weighting — optimizing weights is too easy to overfit on in-sample data.
The Results
On US data 2019-2025, the 5-factor equal-weight ensemble produced: - Sharpe ratio: 0.70 (on 5-day forward returns) - Positive IC across most factors individually - IC robustness: consistent across bootstrap resamples and across 2-year rolling windows
A Sharpe of 0.70 at a 5-day horizon is meaningful. Daily Sharpe statistics are usually noisier — 5-day captures trend continuation without capturing the mean-reversion noise that dominates at 1-day. The 0.70 figure is after equal-weighting all signals, which is a conservative aggregation methodology.
Why Two Systems
The chart system (pattern branch) and the quant system (Simons branch) are designed to have low correlation. Pattern breakout signals cluster at specific market moments — breakout days are discrete events. The Simons ensemble rebalances weekly on factor exposures, which generates smoother return profile with different drawdown timing.
This isn't about one system being better than the other. It's about whether the combined portfolio has better risk-adjusted returns than either alone. The correlation between systems matters more than their individual statistics.
Measured over the 6-year walk-forward period: the correlation between daily PnL of the pattern branch and Simons branch is negative (-0.14 in the most recent measurement). Negative correlation means the systems tend to lose at different times. Portfolio Sharpe benefits from this.
What Jim Simons Would Say
Simons would probably say my ensemble is too small (5 signals vs hundreds), the rebalance frequency is too long (5-day vs intraday in Medallion), and equal-weighting is leaving information on the table.
He'd be right on all three counts. The Simons branch isn't Medallion. It's a systematic, non-discretionary counterpart to my chart work — designed to generate uncorrelated return and to test whether the chart-based approach has edge that pure factors don't capture.
The answer so far: yes, the chart approach captures something that doesn't load on the five factors I've included. The factor IC is 0.70 Sharpe; the pattern branch on US data produces different return timing. This suggests there is orthogonal information in the chart read that the quantitative signals don't summarize.
The Bigger Test
Both systems run paper-trading independently until November 2026. At that point, a branch arbiter (built in June 2026) evaluates the walk-forward results across four statistical criteria: per-trade profit factor, DSR (Deflated Sharpe Ratio), PBO (Probability of Backtest Overfitting), and bootstrap CI. The branch with the strongest results takes the live capital allocation.
The Simons branch might win. It might not. The test is designed to answer the question: given 7 months of live paper trading, which framework produced more robust evidence of edge — discretionary chart reads or pure quantitative signals?
Track. Study. Wait. Strike.
Personal research and trading journal — not investment advice. The author does not provide licensed advisory services. — MOEasymmetry
Draft 2026-06-12. Source: Simons branch ensemble design and backtest 2026. 5-factor equal-weight: RS, price/52w-high, earnings growth slope, volume trend, cross-sectional momentum. US 2019-2025 Sharpe 0.70 on 5d forward returns. Pattern branch vs Simons branch correlation: -0.14. Branch arbiter: overfitting_stats.py + branch_selection_arbiter.py, DSR + PBO + bootstrap. Final branch selection November 2026. Full methodology at project_simons_ensemble_validated.md.