This post does two things. It reports how a portfolio of our eight composite signals performed in October 2026, and it compares the composite signals with the single-model versions they replaced. Each composite signal is built as a committee of Pythia models; each version it replaced was built on one model. All returns are realized, unlevered and net of costs, with daily P&L on fixed capital summed over the month.
Portfolio performance
Table 1 shows the portfolio of the eight composite signals under three weightings: equal weight (1/8 per signal), inverse volatility (weights set from each signal's 2025 volatility), and blended weights, which split the portfolio 50% equal weight and 50% inverse volatility. Each portfolio return links to its report.
Table 1: Composite signal portfolios, September and October 2026
| Portfolio | Sep return | Sep Sharpe | Oct return | Oct Sharpe | Oct max DD |
|---|---|---|---|---|---|
| Equal weight | +0.31% | 1.13 | +0.92% | 3.86 | −0.45% |
| Inverse volatility | +0.28% | 1.34 | +0.88% | 4.98 | −0.28% |
| Blended weights (50/50) | +0.29% | 1.26 | +0.90% | 4.43 | −0.34% |
Realized monthly returns, unlevered, net of costs. Sharpe ratios are annualized. Max DD is maximum drawdown.
Composite signals vs the versions they replaced
Table 2 compares each composite signal with the single-model version it replaced, for September and October. The portfolio row is the equal-weight combination of the eight signals, and each composite return links to the report for that signal and month.
Table 2: Composite vs single-model returns by signal, September and October 2026
| Signal | Sep composite | Sep single | Oct composite | Oct single |
|---|---|---|---|---|
| Portfolio (8 signals, equal weight) | +0.31% | −0.69% | +0.92% | −0.92% |
| SP500_DTC | +0.69% | −1.48% | −0.30% | +0.28% |
| NASDAQ100_DTC | +0.54% | +1.13% | +0.09% | −0.54% |
| RUSSELL2000_DTC | +0.07% | +0.03% | −0.30% | −0.51% |
| DOW30_DTC | +0.67% | −1.28% | +0.05% | +0.09% |
| GOLD_DTC | −0.48% | +0.39% | +3.88% | −1.24% |
| SILVER_DTC | −0.45% | −3.90% | −0.05% | +0.10% |
| OIL_DTC | +1.54% | −3.14% | +5.87% | −7.04% |
| NATGAS_DTC | −0.11% | +2.74% | −1.93% | +1.52% |
Realized monthly returns, unlevered, net of costs. Composite: current signal versions. Single: the single-model versions they replaced. Both months are out of sample for every version shown, except 1 September for the four equity index composites.
What the numbers show
All three portfolio weightings were positive in both months. In October the equal-weight portfolio returned +0.92%, with a Sharpe ratio of 3.86 and a maximum drawdown of −0.45%. The inverse-volatility and blended-weights portfolios returned +0.88% and +0.90%, with smaller drawdowns (−0.28% and −0.34%) and higher Sharpe ratios (4.98 and 4.43).
At the portfolio level, the composite signals outperformed the single-model versions they replaced in both months: +0.92% against −0.92% in October, and +0.31% against −0.69% in September.
Signal by signal, the picture is more mixed. The composite version returned more in 9 of 16 signal-months (5 of 8 in September, 4 of 8 in October). The clearer difference is in the size of the losses: the worst composite signal-month was −1.93% (NATGAS_DTC, October), against −7.04% for the worst single-model signal-month (OIL_DTC, October). The largest October gains came from OIL_DTC (+5.87% against −7.04%) and GOLD_DTC (+3.88% against −1.24%).
Two months is a short record, and we will publish this comparison every month. So far the results are consistent with the reason we moved to composite signals: a committee of models offsets the errors of any single model, and the effect shows most clearly at the portfolio level. We introduced the composite methodology in our Pythia v0.12.0 post.