Public memo · cryptographically signed

NVDA Multiple Risk Favors Regime-Managed AI Exposure

Run 25/5/2026
regime_blend
4 assets
daily rebalance
signed · wRKTmyEG59lkKWt3
Marking position to market…
Sharpe
0.94
Risk-adjusted
Sortino
1.04
Downside-adj.
Ann. return
15.56%
CAGR
Max DD
-17.1%
Peak-to-trough
Vol (ann)
11.8%
Standard dev
CVaR 95%
-1.67%
Tail loss
Win rate
53.3%
Daily positive
Total return
174.3%
Since inception
Equity curve · USD
Portfolio vs. SPY
Portfolio SPY
Selected window
2019-05-28 2026-05-22 · 2551d
Portfolio+174.28%
SPY+195.46%
Alpha-21.17%
Ann. alpha-3.35%
2019-05-28drag a handle · drag the band to slide the window2026-05-22
2019-052021-022022-112024-082026-05
Drag the slider handles to zoom the chart. Click-drag on the chart itself to mark a sub-window and read its gain / alpha.
Multi-model review
Disagreement · 21%

A second frontier model re-reads the same evidence. Thesis stance is what gets signed; the independent row is for audit and risk review — track accuracy over time on Model scorecard.

Primary · Signed
Supported
Evidence 7/10
Independent · OpenAI (gpt-5.5)
Mixed evidence
Evidence 6/10
Evidence gap · Δ -1 vs primary · Independent model is more cautious about evidence strength.
Independent rationale: I would not upgrade this to buy on the evidence shown: the backtest is respectable, but the strategy appears to have earned much of its return from the same semiconductor/AI beta it is supposed to manage, and the excess value versus simply owning SPY is not clearly demonstrated. The regime switch may reduce drawdowns in slow-moving selloffs, but it is poorly suited to the actual stated risk: a sudden NVDA/AI multiple reset that gaps through trend and volatility signals.
Top independent risk: The portfolio is not truly ex-NVDA risk because SMH and AMD remain highly exposed to the same AI-capex narrative, so a sharp NVDA derating could hit the strategy before its regime filter can move to BIL.
Read in:

Research simulation only · Not investment advice · Not a trade recommendation · Past backtests do not predict future results.

Before investing, consult a SEBI-registered research analyst or investment adviser. Attevia is not registered and does not provide personalised advice.

Thesis stance · Supported
Evidence 7/10

Thesis is directionally supported by the simulation, with documented risks. Independent model review is shown separately — it does not change what is signed unless the evidence gate applies.

NVDA Multiple Risk Favors Regime-Managed AI Exposure

NVDA may underperform SPY over the next 12 months despite strong fundamentals because AI-capex perfection is embedded in the sales multiple.
Executive summary

The operating facts remain exceptional: Q1 FY27 revenue was $81.6B, up 85% YoY, with data center at $75.2B and Q2 guidance of $91B ±2%. The issue is valuation fragility, with the stock around $135/share and roughly 28x forward sales while China data center revenue has fallen to zero from $4.6B a year ago. The tested implementation avoids direct NVDA ownership and uses SPY, SMH, AMD and BIL to remain risk-on only when trend and volatility conditions are supportive. The backtest is positive but imperfect: a 0.94 Sharpe and 15.56% annualised return support a buy rating, while semiconductor factor exposure and delayed regime signals limit conviction.

Key findings
  • The strategy delivered a 174.28% total return from 2019-05-28 to 2026-05-22, equivalent to a 15.56% annualised return.
  • Risk-adjusted performance was credible, with an 11.84% annualised volatility, 0.94 Sharpe and 1.04 Sortino.
  • Downside control was material but not complete: maximum drawdown was -17.14% and CVaR 95% was -1.67%.
  • The daily win rate of 53.3% suggests the edge is incremental rather than dominant, consistent with a regime-switching allocation rather than a pure directional call.
  • The original trip-wire remains hyperscaler deceleration: hyperscale revenue was $38B in the quarter, up 12% QoQ versus enterprise/AI cloud up 31% QoQ.
  • The strategy fits the stated late-cycle AI-capex euphoria regime by rotating to BIL when trend and volatility conditions deteriorate.
Risks
  • The implementation can remain exposed to the AI-capex factor it seeks to sidestep because SMH has large NVDA exposure and AMD is tied to the same semiconductor narrative.
  • The 126-day momentum and 21-day volatility signals may react after the first leg of a semiconductor derating, leaving the portfolio exposed to overnight NVDA, AMD or custom-silicon shocks.
  • The 25% annualised volatility cutoff can create whipsaw, particularly if semiconductors rally with elevated realised volatility or decline slowly while volatility remains contained.
  • Daily rebalancing and, for an India-based vehicle, USD/INR exposure introduce execution, tax, hedging and currency risks that are not captured by the headline USD backtest.
Method / sources
Original thesis: NVDA Q1 FY27 revenue, data center revenue, Q2 guidance, forward sales multiple, China data center revenue and hyperscaler trip-wireBacktest: Regime Switch Ex NVDA Core, daily regime_blend allocation, 2019-05-28 to 2026-05-22Critic concerns and stress scenarios provided for Regime Switch Ex NVDA CorearXiv:1603.01580v2 — Cross-response in correlated financial markets: individual stocks
Strategy
Regime Switch Ex NVDA Core
regime_blend
daily

Stays in SPY/SMH/AMD only when trend is positive and vol contained; shifts to BIL when AI-capex euphoria breaks.

Current allocation
SPY
SPDR S&P 500
31.8%
SMH
VanEck Semiconductor
18.2%
AMD
Advanced Micro Devices
14.9%
BIL
SPDR 1-3 Month T-Bill
35.2%
Why this is tamper-proof

Viewing this page, you're trusting Attevia's server. The signature is what matters once this memo leaves it — in a PDF, a screenshot, a pitch deck, or after Attevia is gone. It does three things no screenshot can: it proves these exact numbers haven't been altered since signing (even by us), it proves when the call was made (you can't backdate a track record), and it's verifiable by anyone offline against /api/pubkey without trusting Attevia at all. Change a single digit and the verification breaks. Software, not investment advice.

NVDA Multiple Risk Favors Regime-Managed AI Exposure — Attevia