A Multi-Regime Asset Rotation Framework Incorporating Cryptocurrency as a Risk-On Asset: Empirical Evidence 2016–2025

Abstract

Keywords: systematic rotation, cryptocurrency, regime detection, multi-asset, momentum, risk management, drawdown control

JEL Classification: G11, G12, G17

Risk Disclosure: This paper is for research purposes only. Past performance does not guarantee future results. All simulations use a hindsight-optimal regime map that is not achievable in live trading. See Section 13 for full audit findings.

1. Introduction

The emergence of Bitcoin (BTC) as an institutionally recognised asset class since 2017 presents systematic portfolio managers with both challenge and opportunity. With annualised volatility regularly exceeding 80% and occasional monthly gains above 50%, Bitcoin's return distribution is fundamentally incompatible with classical mean-variance optimisation frameworks that assume Gaussian returns.

Simultaneously, Bitcoin's halving cycle creates a well-documented four-year structural pattern of alternating bull and bear regimes (Panagiotidis et al., 2019). An investor who correctly identifies and participates in Bitcoin bull markets while rotating to defensive assets during bear markets would theoretically capture extraordinary risk-adjusted returns. The central question is whether such identification is achievable systematically in real time.

The Wardell Rotation System (MARS V2) addresses this by implementing a regime-conditional allocation framework: a monthly composite score derived from four independent signal components drives allocation across a spectrum from aggressive (100% BTC) to defensive (100% government bonds), with two intermediate states using equity and gold.

This paper makes three contributions. First, we provide the first rigorous independent evaluation of the MARS V2 strategy using real monthly price data rather than the simulated Gaussian returns used in earlier versions. Second, we conduct institutional-grade robustness testing including execution lag analysis, regime noise sensitivity, Monte Carlo simulation, and stress tests. Third, we provide a frank assessment of the strategy's limitations and the conditions under which live performance would deviate from the historical simulation.

2. Strategy Description

2.1 Composite Regime Score

The composite score C is computed monthly as a weighted linear combination of four normalised signal components:

where each component is normalised to [0, 1]. The components are:

• T (Technical, weight 0.50): Price momentum signals derived from the SPY ETF, including trend direction, RSI, 52-week high proximity, and moving average crossovers.
• N (News Sentiment, weight 0.20): Natural language processing of financial news headlines via NewsAPI. Bullish/bearish keyword frequency ratio, normalised to [0,1].
• M (Macro, weight 0.20): FRED macroeconomic data. Two-month unemployment rate trend, used as a recession indicator. Rising unemployment reduces M; falling unemployment increases M.
• E (Exposure, weight 0.10): Market exposure/volatility proxy. Derived from VIX or equivalent volatility measures.

2.2 Regime Classification

2.3 Portfolio Construction Rules

• Rebalancing frequency: Monthly (first business day)
• Position concentration: 100% single-asset allocation per regime
• Leverage: None (IBKR cash account)
• Transaction costs: £1.70 flat fee for ETF trades; 0.18% rate for BTC; 0.009% effective ETF slippage; 0.10% BTC slippage
• No intra-month stop-loss in S1 base variant

2.4 S6 Combined Variant

An enhanced variant (S6) adds two risk filters: (1) an asymmetric confirmation filter requiring 2-month consensus before upgrading to a higher-risk asset (but allowing immediate downgrade), and (2) a BTC gate that suspends BTC allocation for one month following a single-month BTC return below −15%. These filters reduce whipsawing near regime boundaries and limit BTC tail exposure.

3. Data & Methodology

3.1 Data Sources

All performance simulations in this paper use real monthly price data downloaded from Yahoo Finance via the yfinance Python library, using month-end closing prices. This represents a methodological improvement over earlier versions of this research that used Gaussian-simulated returns parameterised on historical annual means and volatilities.

3.2 Return Computation

Monthly returns are computed as end-of-month price relatives: r_t = P_t/P_t-1 − 1, where P_t is the last trading day closing price in month t. All prices are adjusted for dividends and splits (auto_adjust=True in yfinance).

3.3 Transaction Cost Model

Transaction costs are applied whenever the strategy rotates between assets. The cost model is:

3.4 Deposit Schedule

The simulation models an initial deposit of £1,500 at inception (January 2016), followed by five additional monthly deposits of £500 each (months 2–6), totalling £4,000 invested capital. All subsequent months have zero new deposits — returns are purely from compounding.

3.5 Critical Methodological Note: The Oracle Regime Map

4. Performance Results

4.1 Headline Statistics

Figure 1 — Equity Curves vs Benchmarks

Figure 1: Strategy equity curve (teal) vs SPY buy-and-hold (grey dashed), QQQ (dark grey), Bitcoin buy-and-hold (orange dashed), naive Bitcoin B&H (gold dashed), and 60/40 portfolio (purple dash-dot). Lower panel: underwater drawdown chart. Bottom strip: regime classification timeline (oracle labels). Period: January 2016 – April 2025. Starting capital: £1,500 + £500/month for 5 months.

4.2 Full Performance Attribution Table

* MARS V2 uses hindsight oracle regime map. Live performance will differ. See Section 13.

Figure 5 — Benchmark Comparison

Figure 5: Side-by-side comparison of CAGR (left), Sharpe ratio (centre), and maximum drawdown (right) across MARS V2, SPY, QQQ, BTC buy-and-hold, 60/40, and naive Bitcoin B&H. MARS V2 highlighted in teal.

5. Regime Analysis

5.1 Regime Time Allocation & Return Contribution

5.2 Regime Transition Matrix

Figure 7 — Return Distributions by Regime

Figure 7: Monthly return histograms for each of the four regimes. BULL months (top-left) show a wide right-skewed distribution driven by Bitcoin's bull cycles. BEAR months (bottom-right) show mostly small negative returns from sovereign bonds in a low-yield environment.

6. Risk & Drawdown Analysis

6.1 Drawdown Profile

The strategy's maximum drawdown of -34.0% occurred during the 2021–2022 transition from late bull to bear regime. This compares favourably to Bitcoin buy-and-hold (-75.6%) and represents a 41.6 percentage point reduction in peak-to-trough decline. The Calmar ratio of 2.952 (CAGR / |MaxDD|) indicates favourable return-per-unit-of-drawdown relative to the benchmarks.

6.2 Tail Risk Statistics

Figure 6 — Audit Risk Matrix

Figure 6: Risk matrix plotting identified strategy risks by likelihood of materialising (x-axis) and impact on returns (y-axis). Red zone: critical risks requiring immediate attention. Amber zone: significant risks to monitor. Green zone: managed or minor risks.

7. Rolling Statistics

Rolling 12-month statistics reveal regime-dependent performance clustering. The strategy exhibits periods of exceptionally high rolling Sharpe (>2.0 during BTC bull phases) alternating with near-zero or slightly negative rolling Sharpe during bear/transition periods. This pattern confirms the strategy's returns are not uniformly distributed — they are concentrated in specific market regimes.

Figure 2 — Rolling 12-Month Statistics

Figure 2: Top — Rolling 12-month Sharpe ratio. Green shading indicates positive SR periods; red shading negative SR periods. Centre — Rolling 12-month CAGR. Bottom — Rolling 12-month annualised volatility. All computed using a trailing window of 12 monthly returns.

Figure 3 — Monthly Returns Heatmap

Figure 3: Monthly returns heatmap for the full 2016–2025 period. Green cells indicate positive months; red cells negative. The intensity of colour corresponds to return magnitude (scale: ±70%). Months in BULL/BTC regime (2017, mid-2019, 2020H2, 2021, 2023) display the highest positive values, while defensive months show compressed but mostly positive returns.

8. Monte Carlo Simulation

To quantify outcome uncertainty, we run 2,000 block-bootstrap paths. In each path, monthly returns are drawn in 3-month blocks from the historical sequence (preserving short-term autocorrelation), with small Gaussian perturbations added (σ = 0.5% per month). The same deposit schedule is applied to each path.

Figure 4 — Monte Carlo Simulation Results

Figure 4: Left — histogram of 2,000 final balance outcomes. Right — histogram of CAGR across 2,000 paths. Teal dashed line = actual strategy result; orange = median Monte Carlo path; red dotted = 5th percentile. Block bootstrap method preserves short-term autocorrelation.

9. Robustness & Sensitivity Tests

9.1 Execution Lag Sensitivity

We test the impact of receiving the regime signal N months late, simulating the effect of delayed signal computation or delayed execution.

Figure 8 — Performance vs Execution Lag

Figure 8: CAGR (left), Sharpe ratio (centre), and maximum drawdown (right) as a function of regime signal execution lag (0–6 months). Monotonic performance degradation with increasing lag confirms the strategy's dependence on timely signal generation.

9.2 Regime Label Noise Sensitivity

We test performance when a random fraction of monthly regime labels are replaced with randomly selected regimes, simulating live signal classification errors.

Figure 9 — Performance vs Signal Error Rate

Figure 9: Mean CAGR (solid teal line) and 5th–95th percentile band (shaded) across 50 random noise realisations, plotted against the percentage of monthly regime labels incorrectly classified. The leftmost point (0%) represents oracle performance. Moderate degradation up to ~20% error rate, with increasing variance at higher error rates.

9.3 Fee Sensitivity

The strategy is fee-insensitive due to returns being dominated by BTC bull market gains that far exceed transaction costs. Even with 10× fees the final balance remains above £1.8M.

10. Benchmark Comparison & Correlation

Figure 11 — Correlation Matrix

Figure 11: Pearson correlation matrix of monthly returns between MARS V2 and major asset classes (2016–2025). Values represent linear correlation coefficients. MARS V2's correlation structure shifts with regime — BULL months produce high BTC correlation, while BEAR months produce high bond correlation — giving an overall moderate average correlation with any single asset class.

Figure 12 — Full Performance Table

Figure 12: Institutional performance summary table comparing MARS V2 against S&P 500, QQQ, and Bitcoin buy-and-hold across all key risk-adjusted return metrics.

11. Complete Monthly Record

The following table presents the complete month-by-month strategy record for the full 112-month simulation period. All returns are actual market returns (not simulated). The Regime column shows the hindsight-labelled regime classification; the Asset column shows the allocated holding for that month.

12. Audit Findings

13. Conclusions & Forward Outlook

13.1 Is the Strategy Likely Robust?

The strategy framework is conceptually robust in its defensive rotation logic, which has analogues in established multi-asset momentum and regime-switching literature. However, the specific backtest performance is not robust to the look-ahead bias embedded in the regime map. The key unresolved empirical question — whether the live signal engine can accurately identify BULL regimes in real time — determines whether live performance resembles the oracle backtest or a substantially degraded alternative.

13.2 Realistic Forward CAGR Estimate

13.3 Recommended Development Path

1. Validate the live signal engine by back-calculating C-scores using only data available at each signal generation date. Compare to the oracle regime map and measure classification accuracy.
2. Paper trade for 3+ months before committing real capital. Log every signal and execution to build a forward-validated track record.
3. Implement confidence-weighted position sizing — scale BTC allocation proportionally to C-score distance from regime boundary. Reduces whipsawing and position concentration at uncertain regime boundaries.
4. Engage a tax specialist before first live trade to establish CGT accounting framework for automated crypto trading.
5. Deploy at £1,500–£3,000 initially once paper trading validates signal accuracy. Scale as performance is confirmed.

13.4 Complementary Strategies

The MARS V2 monthly rotation framework operates on monthly timeframes. A complementary grid trading strategy operating on daily timeframes can generate intra-month yield during NEUTRAL and WARNING regime periods when MARS allocates to equity/gold. The two strategies are structurally non-correlated and share a common regime detection layer: when MARS signals a trending BULL regime, the grid bot should pause; when MARS signals NEUTRAL/WARNING (ranging conditions), the grid bot can activate on the allocated asset.

Appendix

A. References

Asness, C., Moskowitz, T., & Pedersen, L. (2013). Value and momentum everywhere. Journal of Finance, 68(3), 929–985.
Brinson, G., Hood, L., & Beebower, G. (1991). Determinants of portfolio performance. Financial Analysts Journal, 47(3), 40–48.
Moskowitz, T., Ooi, Y. H., & Pedersen, L. H. (2012). Time series momentum. Journal of Financial Economics, 104(2), 228–250.
Panagiotidis, T., Stengos, T., & Vravosinos, O. (2019). The effects of markets, uncertainty and search intensity on bitcoin returns. International Review of Financial Analysis, 63, 220–242.
Wardell, R. (2024). A Four-Regime Multi-Asset Rotation System (MARS). SSRN Abstract 6653818.

B. Reproducibility

All results in this paper can be reproduced by running mars_audit.py in the research/ directory of the CPROJECTC project. Data source: Yahoo Finance via yfinance. Random seed: 42. Python 3.10+. Required packages: numpy, pandas, yfinance, scipy, matplotlib.

C. Regime Map (Full Oracle Sequence, 2016–2025)