Objective: Generate consistent, positive risk-adjusted returns with low correlation to traditional stock and bond indices. The strategy harvests carry (income) premia across multiple asset classes while hedging unwanted market exposures using inverse ETFs.
How It Works: Each month, the strategy evaluates 95 candidate "carry streams" — each a long position in a yield-bearing ETF paired with an inverse ETF hedge. The top 5 streams ranked by trailing 1-year Sharpe ratio are selected and weighted proportionally. All positions are long-only ETF purchases — no shorting, no margin, no derivatives.
Asset Classes: Investment-grade bonds, high-yield bonds, emerging market debt, CLOs, MLPs, dividend equities, REITs, commodities, crypto (BTC/ETH/SOL via ETFs), currencies, and managed futures. Inverse treasury and equity ETFs (TBF, TYO, SH) are used as hedges.
| Frequency | Every 21 trading days (approximately monthly) |
| Selection | Rank all 95 streams by trailing 252-day Sharpe. Select top 5 with Sharpe > 0. |
| Weighting | Proportional to trailing Sharpe (capped at 90th percentile to limit concentration) |
| Vol Targeting | Each stream scaled to 3% ann. vol. Portfolio scaled to 10% ann. vol. |
| Risk Controls | VIX stress scaling + exponential drawdown pullback. Both use prior-day data only. |
| Transaction Costs | 5 bps modeled for standard ETFs, 10-15 bps for leveraged/crypto |
eval_data = vol_t.iloc[i-252:i], Python excludes index i) and by autocorrelation test (lag-1 = 0.052, well below 0.1 threshold).Every 21 trading days, the model runs this exact process using only data available at that point. No future information is ever used.
Each "stream" is a daily return series built from one of 10 engines. The two main types:
Carry Streams (no signal — runs every day)
A fixed-weight blend of a yield-bearing ETF + an inverse ETF hedge. Example: carry_hyg_tbf = 67% HYG + 33% TBF every day. The carry (income) from HYG is captured while TBF hedges out interest rate risk. About 70 streams are carry-based.
Momentum Streams (signal-driven)
A position that varies between 0% and 100% based on trailing momentum. The signal averages 21/63/126/252-day risk-adjusted returns. If the signal is positive, go long; if negative, stay in cash. About 40 streams are momentum-based.
Each stream's daily return is scaled so its rolling 63-day volatility equals 3% annualized. This makes all streams comparable — a low-vol CLO stream gets scaled UP, a high-vol crypto stream gets scaled DOWN. The scaler always uses yesterday's volatility (no look-ahead).
Using the last year of vol-targeted returns: Sharpe = mean / std × √252. Only streams with trailing Sharpe > 0.75 pass the quality filter. This ensures only streams with strong recent risk-adjusted performance are considered.
Sort all qualifying streams by their trailing Sharpe ratio. Take the top 5. The number of qualifying streams varies — as of the latest rebalance, 33 of 112 streams had Sharpe > 0.75.
Each position's weight is proportional to its trailing Sharpe squared. This concentrates capital into the highest-conviction stream. If stream A has Sharpe 2.0 and stream B has Sharpe 1.0, A gets 4× the weight of B (not 2×). Weights are capped at the 90th percentile to prevent extreme concentration.
The portfolio earns the weighted sum of its 5 streams' daily returns. No changes until the next rebalance. The carry streams run on autopilot (fixed weights); the momentum streams adjust their position daily based on their own signal, but which streams are selected only changes at rebalance.
Three adjustments are applied to the combined portfolio return each day, all using prior-day data only:
VIX Scaling: If VIX is in its 90th percentile (high fear), reduce exposure to ~50%. If low VIX, allow up to 120%. Formula: scale = 1.2 − 0.8 × VIX_percentile, clipped to [0.5, 1.2].
Drawdown Control: If the portfolio is in a drawdown, reduce exposure exponentially. At −10% drawdown, exposure drops to ~60%. At −20%, it drops to ~37%. Formula: scale = e^(drawdown × 5), clipped to [0.2, 1.0].
Vol Targeting: Scale the entire portfolio so its rolling 63-day volatility equals 10% annualized. If recent vol was 15%, the scaler is 10/15 = 0.67. If vol was 5%, scaler is 10/5 = 2.0.
The backtest applies several adjustments DAILY, not just at monthly rebalance. This materially affects performance.
| What | Frequency | Impact |
|---|---|---|
| Which 5 streams to hold | Monthly (21 days) | Determines your ETF allocation |
| Sharpe² weights between streams | Monthly (21 days) | Determines $ amounts per ETF |
| Per-stream vol scaling | Daily | Scales each stream so its vol stays at 3%. Critical — adds ~0.4 Sharpe. |
| VIX stress scaling | Daily | Reduces total exposure at high VIX. Adds ~0.05 Sharpe. |
| Drawdown control | Daily | Reduces exposure in drawdowns. Adds ~0.04 Sharpe. |
| Portfolio vol targeting | Daily | Scales portfolio to 10% vol. Part of per-stream vol scaling impact. |
Performance gap if you skip daily adjustments:
| Approach | Sharpe | Return | Max DD |
|---|---|---|---|
| Full backtest (daily scaling) | 1.67 | +16.0% | -15.0% |
| Monthly allocation + daily vol target | 1.58 | +16.2% | -20.0% |
| Pure monthly (set and forget) | 1.16 | +8.7% | -19.7% |
Recommendation: At minimum, implement daily per-stream volatility targeting. This is the biggest driver of backtest performance. Without it, expected Sharpe drops from 1.67 to 1.16. The VIX/drawdown scaling adds minimal value (~0.09 Sharpe) and can be skipped. In practice, daily vol targeting means checking each stream's recent volatility and adjusting your dollar exposure so each stream contributes ~3% annualized vol to the portfolio.
The strategy harvests carry premia — the yield income from bonds, dividends, and CLOs — while hedging out the main risk factors (interest rates, equity market) using inverse ETFs. The adaptive selection automatically rotates into whichever carry streams are performing best, and the risk controls cut exposure during crises. The combination of diversified carry + dynamic selection + risk management produces a consistently positive return with controlled drawdowns.