A Strategy for (Micro)Strategy
Following the relationship between MSTR and bitcoin.
Research by Robert Carver. Monitored by Stylus Capital.
Updated after the US close
Methodology v2: synchronized closing observations, next-session rebalancing. Historical results have been rebuilt. Previous series
01The signal today
Reference close · 10 Sept 2026, 16:00 New YorkMSTR is below its model-implied valuation. The model favours MSTR against the other leg, at 42% of its average position size.
$11.3bn below the model
A relative valuation estimate, not a price target.
- MSTR reference close
- $128.56
- Bitcoin reference close
- $77,124
- Signal size
- 42%
- of the model's average position
02What changed?
Price moves above. The model's response below. Both share the same dates.
- MSTR change
- Bitcoin change
- Forecast change
03How has it performed?
Reconstructed strategy returns, with costs shown explicitly.
- Net return
- -9.4%
- Gross return
- -5.7%
- Costs deducted
- 3.68 percentage points
Returns use fixed starting capital. Each selected period includes its first session's P&L and is measured from the preceding close.
| Period | Gross | Net | Net Sharpe |
|---|---|---|---|
| Pre-publication | +5.0% | +2.5% | 0.05 |
| Post-publication | -10.8% | -11.9% | -0.59 |
Costs: 5 bps per leg on one-way turnover; 50 bps annual MSTR borrow while short. Bitcoin financing and other implementation costs are excluded.
04Inside the model
The valuation relationship, the hedge and the risk taken through time.
05An illustrative position
Capital: $100,000
Long MSTR
111 shares$14,289 at reference close
Short bitcoin
0.45 BTC$34,552 at reference close
IBIT proxy: 791 shares
Target for 11 Sept 2026, 16:00 New York. 20% annual volatility target. The hedge is sized to spread risk, so the two dollar amounts differ. Quantities use the reference closes above; they are illustrative, not executable quotes.
06The research behind it
Owning bitcoin and owning MSTR are different exposures. This model asks whether MSTR is expensive or cheap relative to the relationship observed over the preceding year.
How the forecast is formed
A rolling regression estimates MSTR's market capitalization from the dollar value of its bitcoin holdings. The fit uses only earlier observations in the preceding 365 days. The residual is actual market capitalization minus fitted market capitalization. Dividing that residual by actual market capitalization and multiplying by 20 produces the forecast. Negative forecasts favour MSTR; positive forecasts favour bitcoin.
Timing, sizing and costs
Both assets are observed at the US equity session's closing cutoff, including daylight saving and early closes. The signal formed after session t is assumed to rebalance both legs at session t+1's close; those holdings first earn returns from t+1 to t+2. These are hypothetical closing-price fills, not recorded executions. Fundamentals are lagged one session. Spread volatility sets position size against a 20% annual volatility target; forecast magnitude scales it relative to an average absolute forecast of 10. Costs include 5 basis points per leg per trade and an assumed 0.50% annual MSTR borrow charge, not all implementation frictions.
Before and after publication
The paper was published on 5 June 2025. Pre-publication covers sessions before that date; post-publication starts on that date. These two periods do not overlap. Results are reconstructed model returns and are not a live trading account's track record.
Data sources and limitations
MSTR and IBIT adjusted session closes come from Yahoo Finance. Bitcoin uses the last trade in the completed Coinbase BTC-USD hourly candle ending at the same session-close cutoff, not the UTC daily close. These are closing observations, not simultaneous bid/ask quotes. Missing session-close candles stop the update. Data runs through 10 Sept 2026.
Holdings and shares come from StrategyTracker's reconstructed history. A one-session delay does not certify when every historical fundamental was publicly available. Results therefore remain a research reconstruction, not a certified point-in-time trading record. Version 2 replaces the earlier timing convention; subsequent version-2 rows are retained when new sessions are appended.
Replication and historical simulation
Rob's supplied script reproduces the published alpha, beta and approximately 29.65% terminal curve, but its historical rolling function uses the final data date.
Each regression excludes the signal date. Holdings and shares are delayed one trading row. Both target legs, including the signal-time hedge, execute one session later; P&L uses the positions actually held across the following interval. Entry fees are charged even when no position was held before the first trade. Timing and source changes mean this reconstructed curve is not the original paper's curve.
Important disclosures
This material has been prepared by Stylus Capital Partners, LLC (“Stylus”) and is provided for informational and educational purposes only. It is not intended to provide investment advice, nor does it take into account the specific investment objectives, financial situation, or needs of any particular investor.
This material does not constitute an offer to sell, or a solicitation of an offer to buy, any securities or financial instruments. Past performance is not indicative of future results. All investments involve risk, including the risk of a complete loss of capital.
Stylus's actual investment strategies and positions may differ materially from the views or models presented here. Data is believed reliable but is not guaranteed for accuracy, completeness, or timeliness.