Evidence of Alpha in a Momentum-Regime Switching Strategy Applied to Digital Assets
Working Paper — February 2026
Abstract
We present a rigorous statistical evaluation of a systematic long-only strategy that dynamically allocates between Ether (ETH), Bitcoin (BTC), and cash (USDC) on the basis of standardised momentum signals and relative-strength ratios. The strategy is backtested on 1,836 daily observations spanning January 2021 through February 2026. On a risk-adjusted basis, the strategy delivers an annualised Sharpe ratio of 0.92 (Sortino 1.04), compared to 0.50 and 0.54 for passive buy-and-hold ETH and BTC allocations respectively.
We subject these results to parameter sensitivity analysis across 448 configurations, walk-forward out-of-sample testing, circular block bootstrap inference, deflated Sharpe ratio adjustment for multiple testing, regime decomposition, and transaction cost stress testing. We find evidence of a positive risk premium associated with the strategy's timing mechanism: the mean return t-statistic is 2.11 (p = 0.035), and 100% of tested parameter configurations produce positive Sharpe ratios, with 98% outperforming buy-and-hold ETH on a CAGR basis.
However, we also identify material concerns: the Deflated Sharpe Ratio (DSR) does not reach statistical significance after correcting for the 448-trial search space; the strategy underperforms passive benchmarks during sustained bull and sideways regimes; and the maximum drawdown of -49.8% remains severe in absolute terms. We conclude that the strategy exhibits genuine conditional alpha, predominantly sourced from bear-market avoidance, but that the statistical case for unconditional alpha is moderated by multiple-testing considerations and a shorter evaluation window.
1. Introduction
The question of whether systematic trading strategies applied to digital assets generate genuine alpha, or merely capture time-varying risk premia, is of direct relevance to institutional allocators. The digital asset class exhibits high volatility (annualised 60-90% for major tokens), pronounced cyclicality, and structural regime shifts that create both opportunity and hazard for systematic approaches.
This report evaluates a specific strategy that operates exclusively in the spot market across two major digital assets (ETH and BTC) and a stablecoin (USDC). Rather than describe the strategy's construction in detail, we focus on the statistical properties of its return stream and the robustness of its apparent outperformance relative to passive benchmarks.
We structure the analysis around six questions an institutional allocator would ask:
- Is the risk-adjusted performance statistically significant?
- Does the strategy survive out-of-sample testing?
- Is the outperformance robust to parameter choice, or is it the artefact of overfitting?
- How much of the return is alpha versus beta exposure?
- In which market regimes does the strategy add or destroy value?
- How sensitive is the return to transaction cost assumptions?
2. Data and Methodology
2.1 Data
Daily close prices for ETH/USDT and BTC/USDT are sourced from Binance, the largest centralised exchange by volume. The sample spans January 2021 to February 2026 (1,865 daily observations). After applying a warmup window for rolling statistics, 1,836 daily observations remain for simulation.
We acknowledge that the live strategy trades UETH/USDC and UBTC/USDC on Hyperliquid, which are synthetic spot instruments that may exhibit minor pricing deviations from Binance spot. We consider this mismatch immaterial for the purpose of signal evaluation.
2.2 Simulation Assumptions
| Assumption | Value | Conservatism |
|---|---|---|
| Execution price | Daily close | Slightly optimistic |
| Taker fee per leg | 0 bps (gross) | Gross of fees; addressed in Section 7 |
| Slippage | Not modelled | Optimistic; addressed in Section 7 |
| Allocation | 100% to single asset or cash | Reflects live strategy |
| Starting capital | $10,000 | Arbitrary; returns are scale-invariant |
| Signal-to-execution lag | 0 bars (same close) | Addressed in Section 7.2 |
2.3 Benchmark Selection
We benchmark against three passive alternatives:
- Buy-and-hold ETH: Purchase ETH at the first available close, hold throughout.
- Buy-and-hold BTC: Purchase BTC at the first available close, hold throughout.
- Equal-weight portfolio (EW): 50% ETH + 50% BTC, rebalanced daily (no transaction costs applied to the benchmark, which slightly favours it).
3. Headline Performance
3.1 Summary Statistics
| Metric | Strategy | ETH B&H | BTC B&H | EW (50/50) |
|---|---|---|---|---|
| Total Return | 498% | 54% | 108% | 102% |
| CAGR | 42.7% | 9.0% | 15.6% | 15.0% |
| Annualised Volatility | 54.7% | 77.3% | 57.9% | 64.5% |
| Sharpe Ratio | 0.92 | 0.50 | 0.54 | 0.55 |
| Sortino Ratio | 1.04 | — | — | — |
| Calmar Ratio | 0.86 | 0.11 | 0.20 | 0.20 |
| Max Drawdown | -49.8% | -79.3% | -76.6% | -76.3% |
| Max DD Duration | 174 days | — | — | — |
| Skewness (daily) | 0.42 | — | — | — |
| Excess Kurtosis | 13.8 | — | — | — |
The strategy's Sharpe ratio of 0.92 exceeds all passive benchmarks by 0.37–0.42 Sharpe units. The Sortino ratio of 1.04 indicates superior performance per unit of downside risk. The maximum drawdown of -49.8%, while meaningfully lower than buy-and-hold ETH (-79.3%) and BTC (-76.6%), remains substantial and would challenge most institutional risk mandates.
3.2 Annual Returns
| Year | Strategy | ETH | BTC | vs ETH | vs BTC |
|---|---|---|---|---|---|
| 2021 | +155.5% | +166.8% | +34.9% | -11.2% | +120.6% |
| 2022 | -11.2% | -68.2% | -65.3% | +57.1% | +54.2% |
| 2023 | +48.5% | +90.1% | +154.5% | -41.6% | -106.0% |
| 2024 | +18.3% | +41.9% | +111.8% | -23.6% | -93.5% |
| 2025 | +59.4% | -11.6% | -7.3% | +70.9% | +66.7% |
| 2026* | -5.9% | -29.3% | -20.0% | +23.3% | +14.0% |
*2026 is a partial year (Jan 1 – Feb 8).
The annual return series reveals a clear pattern: the strategy underperforms the strongest-performing benchmark during bull markets (2023, 2024) but dramatically outperforms during negative-return years (2022, 2025, 2026). In 2021, results were mixed — the strategy trailed ETH but substantially beat BTC. This asymmetry is the primary source of the strategy's compounding advantage and is consistent with a momentum-crash-avoidance mechanism.
4. Statistical Significance of Excess Returns
4.1 T-test on Mean Returns
The mean daily return is tested against the null hypothesis of zero mean.
| Statistic | Value |
|---|---|
| Mean daily return | 0.14% |
| T-statistic | 2.11 |
| P-value (two-tailed) | 0.035 |
The null hypothesis of zero mean return is rejected at the 5% significance level. This provides evidence that the strategy's mean return is positive, though the result does not reach 1% significance.
4.2 Regression Alpha
We estimate single-factor Jensen's alpha by regressing strategy daily returns on each benchmark:
| Factor | Alpha (annual) | Beta | R² | p-value |
|---|---|---|---|---|
| ETH | 31.9% | 0.50 | 0.50 | < 0.001 |
| BTC | 33.0% | 0.58 | 0.38 | < 0.001 |
| EW (50/50) | 30.4% | 0.59 | 0.49 | < 0.001 |
The annualised alpha against the equal-weight benchmark is 30.4%, highly significant. The beta of 0.59 confirms that the strategy captures approximately 59% of the market's upside while avoiding a substantial portion of the downside.
4.3 Information Ratio
The annualised Information Ratio against the equal-weight benchmark is 0.33. While positive, this is below the 0.5 threshold typically considered indicative of skilled active management in traditional asset classes. However, benchmark comparisons in a three-asset universe may not be directly comparable to traditional equity-market IR expectations.
5. Robustness Testing
5.1 Parameter Sensitivity
To assess whether the strategy's performance is a fragile artefact of a specific parameter combination, we evaluate all 448 combinations across 4 core hyperparameters, spanning a wide range of values for each.
| Metric | Value |
|---|---|
| Configs with positive Sharpe | 448 / 448 (100%) |
| Configs with CAGR > BTC B&H | 363 / 448 (81.0%) |
| Configs with CAGR > ETH B&H | 439 / 448 (98.0%) |
| Configs with Sharpe > BTC B&H | 363 / 448 (81.0%) |
| Median Sharpe | 0.637 |
| Sharpe IQR | [0.557, 0.740] |
| Sharpe 5th–95th percentile | [0.466, 0.862] |
| CAGR range | [2.3%, 50.1%] |
| Base case Sharpe percentile | 96th |
Every single parameter combination tested produces a positive Sharpe ratio, and 98% outperform ETH buy-and-hold on an absolute return basis. This is strong evidence that the strategy's performance is not an artefact of a single lucky parameterisation. Note: sensitivity analysis includes 7 bps transaction costs per leg, making these results conservative relative to gross figures.
The lack of a sharp peak in the parameter space — performance degrades gradually rather than collapsing — is a hallmark of a robust, non-overfit strategy.
5.2 Walk-Forward Out-of-Sample Test
We conduct a rolling walk-forward analysis with a 2-year training window and a 1-year test window. Parameters are held fixed at the base case (no re-optimisation). Transaction costs of 7 bps per leg are included.
| Period | Test Window | OOS Return | OOS Sharpe | Excess vs EW |
|---|---|---|---|---|
| 1 | 2023-01 to 2023-12 | +43.0% | 1.17 | -79.3% |
| 2 | 2024-01 to 2024-12 | +14.0% | 0.52 | -62.8% |
| 3 | 2025-01 to 2025-12 | +55.6% | 1.22 | +65.1% |
| Summary | Value |
|---|---|
| Mean OOS return | +37.6% |
| Mean OOS Sharpe | 0.97 |
| Periods with positive return | 3 / 3 (100%) |
| Periods beating EW benchmark | 1 / 3 (33%) |
5.3 Bootstrap Confidence Intervals
We employ a circular block bootstrap (block size = 20 days, 10,000 replications) to construct a non-parametric 95% confidence interval for the annualised Sharpe ratio.
| Statistic | Value |
|---|---|
| Observed Sharpe (net of 7 bps) | 0.88 |
| Bootstrap mean | 0.88 |
| 95% CI lower bound | 0.02 |
| 95% CI upper bound | 1.79 |
| P(Sharpe ≤ 0) | 2.3% |
The bootstrap is computed on returns net of 7 bps transaction costs per leg. The lower bound of the 95% confidence interval is strictly positive, and the probability of a zero or negative Sharpe under bootstrap resampling is 2.3%. The wider confidence interval reflects the shorter sample period.
5.4 Deflated Sharpe Ratio
Following Harvey & Liu (2015), we compute the Deflated Sharpe Ratio (DSR), which adjusts for the number of trials tested and return non-normality.
| Statistic | Value |
|---|---|
| Observed Sharpe (net of 7 bps) | 0.88 |
| Expected max Sharpe under null (448 trials) | 1.37 |
| Deflated Sharpe Ratio | ≈ 0 |
| Significant at 5%? | No |
The DSR is not significant, indicating that after accounting for the 448-trial search space, the observed Sharpe does not exceed what would be expected by chance. However, three important caveats apply: (1) the 448 configurations are highly correlated, not independent, inflating the penalty; (2) 100% of configurations produce positive Sharpes, which itself rejects the null that the signal is pure noise; (3) the 5-year evaluation window constrains the achievable DSR, as longer samples mechanically produce higher DSR values.
6. Regime Analysis
We classify each trading day into one of three regimes based on the trailing 60-day cumulative return of the equal-weight crypto portfolio:
| Regime | Days | Strategy Ann. | EW Ann. | Excess vs EW | Strategy Sharpe |
|---|---|---|---|---|---|
| Bull | 594 (33%) | +168.3% | +194.9% | -26.6% | 2.44 |
| Bear | 282 (16%) | -13.4% | -227.6% | +214.2% | -0.75 |
| Sideways | 900 (51%) | -24.3% | -5.7% | -18.6% | -0.50 |
Bear markets are the entire source of outperformance. During bear regimes, the strategy loses 13.4% annualised while the EW benchmark loses 227.6% — a difference of 214 percentage points.
Bull markets show approximately 86% upside capture. Sideways markets are the strategy's weakness, where momentum signals generate whipsaws. This regime profile is characteristic of trend-following strategies across all asset classes.
7. Sensitivity Analysis
7.1 Transaction Cost Sensitivity
| Fee (bps/leg) | CAGR | Sharpe | Max DD |
|---|---|---|---|
| 0 (base) | 42.7% | 0.92 | -49.8% |
| 3 | 41.0% | 0.90 | -49.8% |
| 5 | 39.9% | 0.89 | -49.8% |
| 7 | 38.8% | 0.87 | -49.8% |
| 10 | 37.2% | 0.85 | -49.8% |
| 20 | 31.8% | 0.78 | -49.8% |
| 30 | 26.6% | 0.70 | -49.8% |
| 50 | 16.8% | 0.56 | -53.9% |
| 100 | -4.5% | 0.20 | -77.9% |
The strategy outperforms BTC buy-and-hold up to approximately 50 bps per leg, and remains profitable (CAGR > 0) up to nearly 100 bps. Even at 30 bps per leg — a conservative assumption for spot trading — the strategy retains a CAGR of 26.6% and a Sharpe of 0.70.
7.2 Execution Timing Bias
| Execution Timing | CAGR | Sharpe |
|---|---|---|
| Same-bar close | 42.7% | 0.92 |
| Next-bar close | 44.8% | 0.96 |
Next-bar execution actually performs better, eliminating execution timing bias as a concern. The base case is slightly conservative.
8. Return Distribution Properties
The strategy's daily return distribution exhibits positive skewness (0.42) and high excess kurtosis (13.8), consistent with inheriting the fat-tailed distribution of its underlying assets. The positive skewness is a favourable characteristic, indicating a slight right-tail bias in returns.
Autocorrelation
The Ljung-Box test yields Q = 41.2 (p < 0.001), indicating statistically significant serial dependence, though individual lag autocorrelations remain below 0.10 in magnitude.
9. Drawdown Analysis
| Rank | Start | Trough | Recovery | Depth | Duration |
|---|---|---|---|---|---|
| 1 | 2021-05-12 | 2021-05-23 | 2021-11-02 | -49.8% | 174 days |
| 2 | 2024-03-12 | 2024-10-23 | 2025-06-10 | -42.8% | 455 days |
| 3 | 2021-11-09 | 2022-07-14 | 2022-08-11 | -36.7% | 275 days |
| 4 | 2022-08-14 | 2023-01-05 | 2023-03-17 | -33.4% | 215 days |
| 5 | 2023-04-19 | 2023-10-18 | 2024-02-27 | -29.9% | 314 days |
The Calmar ratio of 0.86 compares favourably to ETH (0.11) and BTC (0.20), indicating superior return per unit of peak-to-trough risk.
10. Discussion: Is This Alpha?
10.1 Arguments in Favour
- Statistical significance. Mean return t-stat = 2.11 (p = 0.035). Bootstrap P(Sharpe ≤ 0) = 2.3%.
- Parameter robustness. All 448 configurations produce positive Sharpe ratios. Minimum Sharpe = 0.32.
- Out-of-sample persistence. Mean OOS Sharpe = 0.97 (net of 7 bps fees), which exceeds the gross in-sample Sharpe of 0.92. All 3 out-of-sample periods produce positive returns.
- No execution timing bias. Next-bar execution improves results.
- Transaction cost resilience. Profitable up to approximately 50 bps per leg.
- Economic intuition. Bear-market avoidance via momentum is well-documented across asset classes.
10.2 Caveats
- DSR not significant. After 448-trial adjustment, the DSR does not reach statistical significance.
- Regime dependency. Alpha concentrated in bear markets; sideways markets destroy value.
- Severe drawdowns. -49.8% max DD would challenge institutional mandates.
- Limited OOS windows. Only 3 walk-forward periods due to the 5-year evaluation window.
- Borderline significance. The t-statistic of 2.11 (p = 0.035) clears 5% but not 1% significance.
- Structural market risk. Edge could erode if market microstructure evolves.
10.3 Assessment
| Claim | Confidence |
|---|---|
| Outperforms passive crypto over a full cycle | High |
| Robust to parameter choice | High |
| Persists out-of-sample | Moderate-to-High |
| Specific Sharpe (0.92) precisely estimated | Low-to-Moderate |
| Survives strict multiple-testing | Low (DSR ≈ 0) |
11. Conclusion
This report presents a systematic evaluation of a momentum-regime switching strategy applied to ETH, BTC, and USDC. The strategy delivers an annualised Sharpe ratio of 0.92 over 5 years, exceeding passive benchmarks by 0.37–0.42 Sharpe units. The mean return is statistically significant at 5% (p = 0.035), robust across 448 parameter configurations, and persistent in walk-forward testing (mean OOS Sharpe = 0.97).
The alpha is sourced primarily from bear-market avoidance: the strategy captures approximately 86% of bull-market returns while reducing bear-market losses by over 214 percentage points annualised. This is economically intuitive and consistent with the broader momentum literature.
We conclude that the strategy exhibits genuine conditional alpha, predominantly from bear-market avoidance, but that the statistical case is moderated by the 5-year evaluation window and multiple-testing considerations. As the live track record extends, the statistical case will strengthen or weaken accordingly. The strategy is best deployed as a complement to a diversified portfolio rather than as a stand-alone allocation.
References
- Harvey, C. R., & Liu, Y. (2015). Backtesting. Journal of Portfolio Management, 42(1), 13-28.
- Hurst, B., Ooi, Y. H., & Pedersen, L. H. (2017). A century of evidence on trend-following investing. Journal of Portfolio Management, 44(1), 15-29.
- Jegadeesh, N., & Titman, S. (1993). Returns to buying winners and selling losers. Journal of Finance, 48(1), 65-91.
- Moskowitz, T. J., Ooi, Y. H., & Pedersen, L. H. (2012). Time series momentum. Journal of Financial Economics, 104(2), 228-250.
Appendix: Methodology Notes
Sharpe ratio: Annualised = (mean daily return / std daily return) × √365.25.
Bootstrap: Circular block bootstrap, block size 20, 10,000 replications.
Deflated Sharpe Ratio: Harvey & Liu (2015) with non-normality correction. Trials = 448.
Regime classification: Bull/bear defined as trailing 60-day EW return above +20% / below -20%.
Walk-forward: 2-year train, 1-year test, rolling 1 year. Fixed parameters, no re-optimisation. Includes 7 bps transaction costs.