DARK FOREST CAPITAL MANAGEMENT LP
"13F equity value" = market value of this filer's US-listed long equity positions only. It excludes cash, bonds, non-US and short positions, so it understates a fund's true assets under management — often by a lot.
13F holdings are disclosed ~45 days after quarter-end, and they never reveal when within the quarter a fund actually bought. So any 13F-based summary is structurally late and blurred — this applies to every fund, including this one.
We backtested copying it anyway. Buying this fund's new positions the day each filing went public, over 17 quarters, returned +0.7% per quarter — versus +0.8% per quarter from simply owning every 13F stock. It beat that baseline in only 35.3% of quarters (excess t = -1.29, not statistically significant). Its filings tell you what it bought — not what you should buy.
Quarterly compounding, invested quarters only · entry 47 days after quarter-end (when 13F data becomes public)
Top 20 holdings of 766 · 2026 Q1
| Ticker | Value | Weight | QoQ |
|---|---|---|---|
| BPOP | $12M | | ADD |
| AXS | $11M | | TRIM |
| CDE | $11M | | ADD |
| ROKU | $10M | | ADD |
| UNP | $9M | | TRIM |
| TOST | $9M | | TRIM |
| TPG | $9M | | ADD |
| SSNC | $8M | | ADD |
| HUBS | $8M | | ADD |
| ZM | $8M | | ADD |
| MU | $7M | | ADD |
| SOFI | $7M | | ADD |
| AGNC | $7M | | TRIM |
| GOOG | $7M | | TRIM |
| BFAM | $7M | | TRIM |
| AFRM | $7M | | ADD |
| ATMU | $7M | | TRIM |
| FERG | $7M | | ADD |
| AAPL | $7M | | HOLD |
| FCNCA | $7M | | ADD |
QoQ vs previous quarter's share count · NEW = new position · ADD/TRIM = ±2% shares · HOLD = unchanged.
New positions in 2026 Q1
Method & Limitations
Method: a "new position" = held this quarter, absent last quarter (options excluded; stocks with <50 prior holders excluded to filter spin-off artifacts). Entry 47 days after quarter-end — the first day the public could act on the filing. Benchmark = equal-weighted universe of all 13F-held stocks. Limitations: quarterly snapshots can't see intra-quarter trades; survivorship bias — funds that shut down are absent, which flatters the sample. Statistics, not advice.