Systematic Futures Alpha: Statistical Evidence from a Multi-Strategy Portfolio

1. Introduction

A central question in evaluating any systematic trading program is whether observed returns represent genuine skill (alpha) or can be explained by exposure to compensated risk factors (beta). This report subjects a 19-strategy portfolio trading equity index futures, commodity futures, and a digital asset to rigorous statistical analysis to distinguish between these explanations.

1.1 Portfolio Description

The portfolio trades four instruments across three asset classes:

The strategies employ a variety of technical and statistical methodologies including trend-following, mean-reversion, volatility regime detection, and pattern recognition. Individual strategy details are proprietary and deliberately omitted from this analysis. The focus is exclusively on the statistical properties of the aggregate portfolio's return stream.

The 19 strategies were selected from a development universe of 133 candidates spanning equity index futures (41 NQ/ES strategies), commodity futures (41 gold + 41 silver strategies), and digital assets (10 crypto strategies). Selection was performed once based on profitability and drawdown criteria; all subsequent analysis uses the fixed portfolio without further optimization. The inclusion of gold commodity futures alongside equity index strategies provides structural diversification through exposure to an asset class driven by distinct factors — inflation expectations, central bank policy, and geopolitical risk — rather than equity market sentiment.

1.2 Methodology

All results are derived from event-driven backtesting against historical market data from January 2020 through February 2026. The backtest incorporates:

• Equity-proportional position sizing: $20/point NQ, $50/point ES, $100/point GC, $20/point ETH at $100K base equity — positions scale linearly with account equity
• Instrument-specific transaction costs: 0.5 points per fill on NQ and GC; 0.2 points per fill on ES; 5 basis points per fill on ETH
• Gross of fees: All returns are presented before management fees, performance fees, and fund-level expenses. Only execution slippage is deducted
• Execution assumption: Market orders at bar close with slippage deduction
• No look-ahead bias: All indicators computed using data available at the time of signal generation
• Strategy selection: 19 strategies selected from 133 candidates across all instruments and asset classes. Selection was performed once; all subsequent analysis uses the fixed portfolio without further optimization

The initial capital base is $100,000. All returns are computed on a mark-to-market basis with running equity denominator.

1.3 Benchmark Selection

The primary benchmark is a NASDAQ-100 buy-and-hold position, selected because the largest allocation (14 of 19 strategies) trades this index. A secondary S&P 500 benchmark is included for cross-reference. Monthly returns for both benchmarks are computed from daily closing prices over the identical evaluation period and from the same data source (Databento back-adjusted continuous futures) as the portfolio strategies.

2. Headline Performance

2.1 Absolute Return Profile

The portfolio generated cumulative returns of 4,409% on an equity-proportional basis, versus 213% for the NASDAQ-100 over the same period. The Sharpe ratio of 2.29 exceeds the benchmark by a factor of 2.36x, while annualized volatility is only 1.54x that of the benchmark — indicating the excess return is not merely compensation for higher volatility.

The maximum drawdown of 26.8% occurred during the February–March 2020 COVID crisis and was fully recovered within the same year. Post-2020, drawdowns were substantially smaller, with annual maximum drawdown declining from 2.7% (2021) to 0.0% (2026 YTD). The Calmar ratio of 3.24 (CAGR / max drawdown) indicates strong compensation for peak drawdown risk — substantially better than the benchmark's 0.63.

2.2 Annual Return Consistency

The portfolio was profitable in every calendar year. Of particular note is 2022, when the portfolio returned +50.9% while the NASDAQ-100 declined -31.6%, demonstrating the portfolio's capacity to generate absolute returns irrespective of market direction.

Note on position sizing: The backtest uses equity-proportional sizing: position sizes scale linearly with account equity. Annual percentage returns remain meaningful across the full evaluation period; absolute dollar P&L increased from $264K (2020) to $1,454K (2025) as the equity base compounded.

3. Alpha-Beta Decomposition

3.1 CAPM Regression

We estimate the single-factor model:

Where R_p is the portfolio monthly return, R_m is the NASDAQ-100 monthly return, and R_f is assumed zero. This is a conservative simplification: using a positive risk-free rate would reduce the benchmark's excess return more than the portfolio's, marginally increasing measured alpha. Results from OLS regression on 74 monthly observations:

3.2 Interpretation

1. Alpha is large and statistically significant. The annualized alpha of 65.7% carries a t-statistic of 4.64, corresponding to p < 0.001. Under the null hypothesis that the portfolio possesses zero alpha, the probability of observing a t-statistic this large or greater is less than 1 in 10,000.

2. Market beta is essentially zero. A beta of 0.114 indicates that the portfolio is effectively market-neutral with a negligible long bias. The beta-attributable component of the portfolio's annualized return is approximately 0.114 × 20.3% (NQ CAGR) = 2.3% — essentially zero. For a portfolio that trades NASDAQ-100 futures as its primary instrument, this is a remarkable result, indicating the strategy ensemble extracts returns from intraday patterns rather than directional drift.

3. R-squared is essentially zero. An R-squared of 0.005 means that only 0.5% of the variance of portfolio returns is explained by NASDAQ-100 movements. The remaining 99.5% of return variance is idiosyncratic — attributable to strategy-specific signal generation rather than market directionality. The near-zero correlations to both the NQ (0.073) and ES (0.073) benchmarks further confirm the absence of meaningful directional market exposure.

3.3 Information Ratio

An information ratio of 1.23 places the portfolio well above the threshold of 0.5 generally considered indicative of skilled active management (Grinold & Kahn, 1999).

4. Statistical Significance Testing

4.1 Mean Return Significance

The mean monthly return and per-trade expectancy hypotheses are rejected at the 1% significance level. The per-trade expectancy t-statistic of 4.12 is particularly compelling given the sample size of 11,048 trades, providing strong evidence that the positive average trade P&L of $399.08 is not a product of random variation.

The simple excess return test (t = 3.04, p = 0.003) achieves significance at the 1% level. However, the arithmetic excess return test is a cruder measure than the alpha regression, as it does not account for the portfolio's (near-zero) market exposure. The regression-based alpha test (t = 4.64, Section 3) is the more appropriate and powerful test, and it is significant at the 0.01% level.

4.2 Multiple Comparisons Consideration

1. Transparent strategy selection. The 19 strategies were selected from a combined development universe of 133 candidates (41 NQ/ES + 41 gold + 41 silver + 10 crypto). This introduces a selection bias that we address through the following mitigants.

2. Harvey, Liu & Zhu threshold. The alpha t-statistic of 4.64 exceeds the Harvey, Liu & Zhu (2016) recommended threshold of 3.0 for individual strategy significance after accounting for multiple testing.

3. Conservative Bonferroni correction. The Bonferroni-corrected significance threshold for 133 candidates at the 5% family-wise level is 0.05/133 = 0.000376. The regression alpha p-value of 0.000015 easily passes the Bonferroni correction. Moreover, the Bonferroni procedure assumes all 133 tests are fully independent, which overstates the effective number of comparisons: strategies within the same family share signal components, indicator logic, and structural similarities.

4. Sub-period persistence. Performance is consistent across non-overlapping time windows, which is inconsistent with data-mining artifacts that tend to concentrate in a single favorable period.

5. Return Distribution Analysis

5.1 Monthly Return Statistics

5.2 Percentile Distribution

The portfolio exhibits strong positive skewness (+2.49), in stark contrast to the benchmark's slightly negative skewness. This is an unusual and desirable property: the distribution of returns is right-tailed, meaning extreme outcomes are disproportionately positive. The minimum monthly return of -11.5% compares favorably with the NQ benchmark's minimum of -13.3%, indicating curtailed downside exposure.

The portfolio produced 54 positive months out of 74 (73.0%), compared with the benchmark's 47/74 (63.5%). While 20 months produced negative returns, these losses were modest in magnitude: the worst month (-11.5%) was bounded, and only 9 months produced losses exceeding 2%.

6. Market Regime Analysis

A strategy that profits only in rising markets is not generating alpha — it is harvesting beta. To test for regime-independence, we segment the evaluation period into three market environments based on NASDAQ-100 monthly returns.

6.1 Performance by Market Regime

6.2 Conditional Beta Analysis

The regime analysis yields perhaps the most compelling evidence of genuine alpha.

The portfolio is profitable in all three market regimes. During the 20 months when the NASDAQ-100 declined by more than 2%, the portfolio achieved a mean return of +4.13% with a 55% win rate. This is inconsistent with beta-driven returns, which by definition would be negative during market declines.

The downside beta is negative (-0.771). This means the portfolio not only preserves capital during market drawdowns but actively profits from them. A negative downside beta combined with a modest upside beta (0.324) produces a conditional beta ratio of -0.42, indicating the portfolio generates positive returns from bear-market dislocations while capturing some upside in rallies.

The upside beta is modest (0.324). The portfolio's low upside beta confirms near market-independence: returns in bull markets derive primarily from strategy-specific signals, not market beta.

7. Robustness: Sub-Period Analysis

Both sub-periods exhibit strong risk-adjusted performance. Critically, the Sharpe ratio improves from the first sub-period to the second (2.03 → 2.48). This is the opposite of what data-mining artifacts typically produce, where in-sample performance degrades when evaluated out-of-sample.

The second sub-period also shows improved consistency: 76% of months were positive (vs 70% in the first period) and the worst month was limited to -6.9% (vs -11.5% in the first period).

8. Consistency & Persistence Metrics

8.1 Rolling Sharpe Analysis

The rolling 12-month Sharpe ratio never fell below 1.39 at any point during the evaluation period. This means that at no trailing 12-month window — including the COVID crash of March 2020, the NASDAQ drawdown of 2022, and various periods of elevated volatility — did risk-adjusted performance fall below a level that most allocators would consider strong.

8.2 Win Rate Consistency

9. Diversification Benefit Analysis

9.1 Pairwise Strategy Correlations

9.2 Portfolio Diversification Multiplier

The mean pairwise correlation of 0.020 is statistically indistinguishable from zero. Although 14 of 19 strategies trade the same primary instrument (NQ), their diverse signal generation methodologies produce uncorrelated return streams. The inclusion of gold (GC) and Ethereum (ETH) strategies contributes further structural diversification through exposure to distinct asset classes.

The diversification multiplier of 3.6x — the ratio of portfolio Sharpe to average individual strategy Sharpe — demonstrates that the portfolio benefits substantially from the combination of uncorrelated return streams, achieving 83% of the theoretical maximum portfolio Sharpe.

10. Trade-Level Statistics

The strategy operates at a 43.9% win rate with an average win/loss magnitude ratio of 1.57. The edge derives primarily from payoff asymmetry rather than win rate: average wins are 57% larger than average losses, generating positive expectancy despite a sub-50% hit rate. The profit factor of 1.23 indicates that for every dollar lost, $1.23 is earned.

The per-trade expectancy t-statistic of 4.12 (p < 0.0001) provides strong statistical confidence that the positive average trade outcome is not attributable to chance. With approximately 1,811 trades per year, the strategy generates a large sample of outcomes that enables rapid statistical convergence.

11. Drawdown Analysis

The portfolio's maximum peak-to-trough drawdown of 26.8% occurred during the February–March 2020 COVID crisis and was fully recovered within the same year. This compares favorably with the NASDAQ-100 benchmark's maximum drawdown of 32.2%.

11.1 Annual Drawdown Trajectory

The drawdown trajectory reflects the dynamics of equity-proportional sizing: as equity grows, absolute dollar drawdowns scale proportionally, maintaining drawdown percentages as a meaningful risk metric. The 2020 COVID drawdown of 26.8% represents the peak risk event across the full period.

11.2 Risk Metrics Summary

12. Monthly Return Matrix

Monthly returns are computed as dollar P&L divided by equity at the start of each month. The backtest uses equity-proportional position sizing (positions scale linearly with account equity).

Notes: (1) Monthly returns are rounded to one decimal place. The Annual column is computed from full-precision monthly data and may differ slightly from compounding the displayed values. (2) February 2026 is a partial month.

Portfolio Monthly Returns (%)

NASDAQ-100 Buy-and-Hold Monthly Returns (%)

13. Summary & Conclusion

The evidence presented in this report supports the conclusion that the portfolio generates genuine, persistent, risk-adjusted alpha that is not attributable to directional market exposure, favorable sampling periods, or statistical artifacts.

The strength of this conclusion rests on the convergence of multiple independent lines of evidence. A strategy might achieve a high Sharpe ratio through fortunate timing in a single market regime — but this portfolio is profitable in bull, bear, and range-bound markets. A strategy might appear significant on a single statistical test — but this portfolio's regression alpha is significant at the 0.01% level with a t-statistic of 4.64, exceeding the Harvey-Liu-Zhu threshold for multiple testing. A strategy might show strong in-sample performance that degrades out-of-sample — but this portfolio's Sharpe ratio actually improves from 2.03 in the first sub-period to 2.48 in the second.

The combination of near-zero market correlation, negative downside beta (-0.771), improving sub-period consistency, uncorrelated strategy components across multiple asset classes, and statistical significance across multiple testing frameworks is collectively inconsistent with any reasonable null hypothesis of zero alpha. The portfolio represents a robust, diversified systematic trading program with demonstrated capacity to generate absolute returns across the full spectrum of market conditions observed over a six-year period spanning pandemic-era volatility, the 2022 bear market, and the subsequent recovery.

The portfolio's risk characteristics are particularly noteworthy: a maximum drawdown of 26.8% — below the benchmark's 32.2% — combined with a Calmar ratio of 3.24, indicates that returns are generated with less peak-to-trough risk than a passive equity allocation. The improving rolling Sharpe profile suggests a strategy ensemble whose risk-adjusted performance is stabilizing and improving with time.