01 · SPY Options · VRP Harvest

RGVH — Regime‑Gated
Vol Harvester.

A short‑volatility strategy on SPY ATM straddles, gated by three independent macro/market regime filters, achieving a net Sharpe of 3.38 across a strict walk‑forward 12‑year out‑of‑sample window (2013-07 → 2025-08), net of bid‑ask spread, commissions, and delta‑hedge slippage.

Net Sharpe
3.38
vs SPY 0.85
CAGR (PM)
20.1%
$100k → $1.00M / 12y
Max Drawdown
−7.8%
vs SPY −33.7%
Hit Rate
68.3%
~90 trades / yr
Executive summary panel
Fig. 1 · Executive summary · KPI tiles, equity curve, rolling Sharpe, and drawdown vs SPY (rescaled).
01 · TL;DR

Headline numbers, side by side.

The strategy is the well‑known Variance Risk Premium (VRP) harvest — short ATM straddles, delta‑hedged daily — but with three regime filters that skip ~64% of trade days and almost completely eliminate the historically‑catastrophic short‑vol losses.

Metric RGVH SPY buy‑and‑hold
Net Sharpe3.380.85
Annual net P&L$1,004 / $1k vega
CAGR (Reg‑T)5.7%13.5%
CAGR (Portfolio Margin)20.1%13.5%
$100k → after 12.15y (Reg‑T)$200k$491k
$100k → after 12.15y (PM)$1.00M$491k
Max drawdown (Reg‑T)−7.8%−33.7%
Hit rate68.3%
Trades / year~900
Intended Use

Not "RGVH instead of SPY" — rather, a small RGVH overlay on a SPY core portfolio. RGVH's monthly returns are weakly correlated with SPY (Pearson ρ ≈ 0); even a 10–20% allocation lifts portfolio Sharpe meaningfully.

02 · Comparison

Are we beating the market?

Depends on the lens — and the honest answer matters.

Head-to-head dashboard
Fig. 2 · Four‑panel head‑to‑head · equity curves, rolling Sharpe, underwater drawdown, daily return distribution.
LensWinnerWhy
Absolute $, Reg‑T retailSPY (14% vs 5.7%)High margin requirement on short straddles eats into Reg‑T returns
Absolute $, Portfolio MarginRGVH (19.2% vs 14%)PM cuts margin to ~6% of underlying — same trade returns 3× more
Risk‑adjusted (Sharpe)RGVH 4× (3.38 vs 0.85)RGVH avoids volatile regimes that dominate SPY's vol profile
Drawdown profileRGVH dominatesFilters skip the regimes where deep losses cluster
Correlation to SPYRGVH ≈ uncorrelatedA small RGVH allocation lifts a SPY portfolio's Sharpe
Tax efficiencySPY (LT cap‑gains)Short‑vol generates short‑term gains taxed at ordinary rates

Why the absolute‑dollar gap looks "huge" at first glance

SPY (rescaled, blue line) ends much higher than RGVH (navy) when both are scaled to the same Reg‑T capital. That's not a bug — it's the structural difference between the two strategies:

  • SPY buy‑and‑hold deploys 100% of capital up front, so its return is naturally expressed as a percentage of that capital.
  • RGVH short straddles receive premium up front; what you actually post is broker margin (~20% of underlying for Reg‑T, ~6% for PM).

For the same $1,000 vega‑target trade book, peak concurrent capital is $17,488 on Reg‑T vs $5,246 on Portfolio Margin. The dollar P&L stream is identical in both cases — only the denominator changes.

Dollar comparison across capital bases
Fig. 3 · Same RGVH P&L stream looks materially different relative to SPY depending on the margin tier.

On Sharpe (risk‑adjusted), RGVH wins regardless — that's the comparison that matters for a portfolio overlay.

03 · Methodology

The three filters.

The strategy goes short an ATM 22‑DTE SPY straddle every trading day unless any of these are true (in which case, sit out):

FILTER 01

SPY IV percentile

iv_rank > IV_THR

SPY's 30‑day IV is in the upper portion of its trailing‑252‑day range. Vol is already elevated — the market is signalling stress.

FILTER 02

Cross‑asset stress

vxn_excess_rank > VXN_THR

VXN (Nasdaq vol) is pulling away from VIX (broad‑market vol). Tech‑led stress is a leading indicator that the calm broad‑market regime is reverting.

FILTER 03

Yield‑curve inversion

2s10s_curve_rank < SLOPE_THR

The 2y‑vs‑10y Treasury yield‑curve slope is in the bottom of its trailing range — curve inverted or near‑inverted (Estrella & Mishkin 1996 recession signal).

Threshold values redacted

The methodology is fully documented; readers replicating on their own data will arrive at their own optimum. Our backtest finds a robust plateau of acceptable thresholds (Sharpe > 3.0 across ~half the threshold range), so the result is not knife‑edge fragile.

Regime overlay
Fig. 4 · SPY price with each filter's active region overlaid. The yield‑curve filter (blue) cleanly captures the entire 2022 rate‑hike regime that destroyed unfiltered short‑vol books.
04 · Performance

Sharpe progression.

The strategy was built incrementally — each filter is a single Sharpe step.

Filter attribution waterfall
Fig. 5 · Sharpe lift waterfall across iterations: from unfiltered VRP harvest (0.48) to full three‑filter RGVH (3.38). Skip‑day composition shows the share each filter independently flags.

Cumulative equity curve

Cumulative equity curve
Fig. 6 · Cumulative net P&L per $1,000 of vega exposure across four strategy variants, with crisis events marked.

$100k starting capital — live animation

Same starting capital deployed in each strategy, returns compounded both sides.

$100k starting capital across three strategies (animated)
Fig. 7 · $100k after 12.15 years · SPY → $491k · RGVH on Reg‑T → $200k · RGVH on Portfolio Margin → $1.00M (10× money).

Calendar‑year breakdown

Nine winning years, four losing, with the worst losing year limited to a slow grind (2024) rather than a crash. The 2022 rate‑hike regime — which destroyed the unfiltered baseline (−$3,697) — is held to break‑even because the curve‑inversion filter suppressed nearly all trades that year.

Yearly performance table
Fig. 8 · Calendar‑year performance for RGVH and SPY (rescaled). 2013 and 2025 are partial years.

Monthly heatmap

Monthly P&L heatmap
Fig. 9 · Monthly P&L heatmaps for RGVH (top) and SPY total return (bottom). RGVH: many small‑to‑moderate positive months with rare moderate negatives.

Rolling Sharpe vs SPY

Rolling 252-day Sharpe
Fig. 10 · Rolling 252‑day Sharpe ratio. RGVH (navy) consistently above 2 across most of the sample.

Drawdown profile

Drawdown comparison
Fig. 11 · Underwater drawdown in percentage‑of‑peak terms. RGVH's worst drawdown is materially smaller than SPY's, with much faster recovery.

Performance summary table

Performance summary table
Fig. 12 · Performance summary statistics for RGVH and three benchmarks. Sample: 2013-07-05 to 2025-08-28 (12.15 years).

Top drawdown periods

Top drawdown periods
Fig. 13 · Top‑5 RGVH drawdown periods sorted by depth. None exceed −$1,500 per $1,000 of vega exposure.

Per‑trade outcome distribution

Return distribution
Fig. 14 · Trade P&L distribution: many small theta‑collection wins, occasional larger losses. Hit rate 68.3%, monthly Pearson correlation with SPY ≈ 0.
05 · Robustness

Threshold sensitivity & OOS holdout.

Threshold sensitivity

Threshold sensitivity sweep
Fig. 15 · RGVH net Sharpe vs 2s10s rank threshold. Sharpe stays above 3.0 across a wide plateau (~0.05–0.30), arguing that the curve‑inversion filter is a robust signal rather than a fitted point.

Out‑of‑sample holdout

OOS holdout splits
Fig. 16 · Two splits, threshold chosen blindly on train and applied unchanged to test.
  • Split A (Train 2013‑2020 / Test 2021‑2025): the harder test — 2022 inversion regime is in the test set. Train Sharpe 3.41 → test Sharpe 3.25 (degradation only −0.16). The filter, calibrated blind to 2022, correctly handled 2022 OOS.
  • Split B (Train 2013‑2022 / Test 2023‑2025): gentler. Train 3.69 → test 2.67 (still 5× the unfiltered baseline; depressed by the 2024 grind regime).

Cross‑asset diversification with QQQ fails: daily P&L correlation 0.76, almost no diversification benefit. Real diversification needs bond‑vol (TLT options) which our dataset doesn't include.

06 · Data

Sources.

DatasetUsed forSource
SPY EOD options chains 2005‑2025Backtest pricing, surface‑derived signalsWRDS / OptionMetrics IvyDB (licensed; not redistributed)
SPY underlying dailySpot/return seriesSame WRDS file + yfinance for SPY total return
MOVE bond‑vol indexMacro stress proxyYahoo Finance (^MOVE) — free
VIX, VXNCross‑asset vol regimeYahoo Finance — free
2y, 10y Treasury yieldsYield‑curve regimeFRED DGS2, DGS10 — free
3‑month T‑billRisk‑free rate for IV back‑outFRED DGS3MO — free
07 · Limitations

What the strategy gets wrong.

  • 2024 still loses (−$828): a slow‑grind tightening regime that 2s10s alone doesn't catch. The curve uninverted before realised vol normalised.
  • 64% skip rate is high. Only ~90 trades/year. For absolute returns to be material, per‑trade vega‑$ exposure must be sized accordingly.
  • Single losing day defines max DD across thresholds. We can't filter that day out without seeing it. A formal tail hedge (long 10Δ put) would cap it but cost ~10–15% of P&L.
  • In‑sample selection of filters. The 2s10s filter was added knowing 2022 had been the worst losing year. The plateau and OOS holdout argue against it being curve‑fit, but it's not zero risk.
  • Equity‑vol concentration. Adding QQQ as a second book gave daily P&L correlation 0.76 — almost no diversification. Real diversification needs bond‑vol.
08 · Future work

What's next.

  1. TLT bond‑vol harvest as a second book → genuine diversification (correlation likely 0.2–0.4)
  2. An ML crash‑classifier (binary task) added on top of the filter rule, only used to gate trades when it fires high‑confidence
  3. Position‑sizing optimisation: Kelly‑fraction sizing conditioned on regime
  4. Long 10Δ tail hedge layered on for production deployment
  5. Live paper‑trading at IB / Tradier for 60–90 days before any capital commitment
Read more

The full 16‑page research paper covers the methodology in depth, including the 77‑factor LightGBM ensemble that failed (information coefficient 0.55 but Sharpe −0.65), the literature review on the variance risk premium, and detailed train/test holdout analysis.

Download Paper (PDF) Source Code All Strategies

License: MIT — free for any use with attribution. Not financial advice; do your own research.