How to Read a 13F Filing: The Conviction Tier Framework
Position sizes tell you what a fund owns. Quarter-over-quarter changes reveal conviction. A five-tier framework for reading 13F filings like a working analyst.
TL;DR
- Position sizes tell you what a fund owns. Quarter-over-quarter changes tell you what it believes.
- Five conviction tiers: unchanged (highest signal), trimmed-but-kept, adding, fresh entries, exits
- Unchanged positions during a net-selling quarter are the strongest conviction signal in any 13F
- Theme mapping across holdings reveals concentrated bets disguised as diversification (e.g., 40%+ in AI infrastructure)
- The Meta exit after 13 consecutive quarters as #1 position reveals more about fund psychology than any current holding
The Conviction Tier Framework
Most people read a 13F for the big names and position sizes. That is like reading a novel by skimming the chapter titles. The signal lives in the changes between quarters: what moved, what stayed, and what disappeared. A $30 billion fund’s Q4 2025 filing demonstrates every pattern worth knowing.
Position size tells you what a fund owns. Quarter-over-quarter change tells you what a fund believes. The framework that separates casual 13F readers from working analysts is conviction tier mapping.
Tier 1: Unchanged positions. These are the highest conviction signal in any 13F. When a fund is net selling $5 billion across the portfolio and trimming nearly everything, the positions that remain untouched are the ones management refuses to touch. In Q4 2025, this fund held Alphabet at $3.3 billion (11.2% of portfolio), Taiwan Semi at $1.3 billion, and Spotify at $881 million without changing a single share. In a quarter where they reduced exposure by billions, “unchanged” is not passive. It is a deliberate decision to hold.
Tier 2: Trimmed but kept. This is valuation discipline, not a loss of conviction. Microsoft was trimmed by 1.1 million shares (roughly 9%). Amazon by 1.0 million shares (also about 9%). NVIDIA by 698,000 shares. All three remained in the top five positions worth $2.1 billion or more. The pattern: take profits at highs, maintain the thesis. When a fund trims a $2.6 billion position by 9%, they still have $2.6 billion of conviction. The trim is noise. The hold is the signal.
Tier 3: Adding. New money flowing into existing positions reveals where the fund sees the best forward return per dollar. Coupang received a 16% increase. Corpay got 17% more. Broadcom added 6%. These are not new ideas. They are existing theses the fund is doubling down on. When you see additions in a quarter where the fund is a net seller, pay close attention. That capital had to come from somewhere.
Tier 4: Fresh entries. New positions reveal what the fund has been researching for the past several months. Netflix appeared at $242 million. Block was increased from a small base. Each entry represents weeks or months of due diligence that just crossed the conviction threshold. The size tells you how much confidence exists. Netflix at $242 million in a $29.7 billion portfolio is a starter position, roughly 0.8%. The fund is interested, not yet committed.
Tier 5: Exits and heavy reductions. This is where the anti-conviction signals live. Meta was cut 63% in Q3 and taken to near zero in Q4. That was the fund’s number one position for 13 consecutive quarters. Eli Lilly exited completely. CrowdStrike exited completely after the 2024 outage. MongoDB was bought in Q3 ($106 million) and sold entirely in Q4. The Meta exit is especially telling: the fund concluded that $100 billion or more in AI capital expenditure lacked visible returns. They left. When you see a multi-year conviction holding get liquidated in two quarters, something fundamental changed in the thesis.
Cross-Quarter Comparison: Where the Story Lives
A single quarter’s 13F is a snapshot. Two quarters reveal direction. Three or more reveal strategy.
This fund went from over 200 positions at its peak to 54 in Q4 2025. That compression alone tells a complete story: the fund experienced a catastrophic year (down 56% in 2022, $42 billion lost from overextension), rebuilt its approach around concentration, and now runs with surgical precision. The top 10 positions represent roughly 63% of the portfolio. The top five hold about 42%.
Reddit provides another cross-quarter story. The fund reduced its position by 24% in Q3 and another 18% in Q4. That is accelerating exits. The data licensing thesis that originally drove the investment may be losing its edge. You do not need a research note to see this. The 13F tells you directly.
Theme Mapping: What the Portfolio Says About the Future
Individual positions tell you about companies. Portfolio composition tells you about themes.
Map every holding to its investment thesis and patterns emerge:
AI Infrastructure (40%+ of portfolio): Alphabet, Microsoft, Amazon, NVIDIA, Taiwan Semi, Broadcom, Lam Research. Combined value: $13.9 billion. This fund is not betting on AI. They are betting on the physical infrastructure that makes AI run. Compute, cloud, chips, and the equipment that manufactures chips.
Southeast Asia Digital (10%): Sea Limited, Grab, Coupang. Combined: $3.1 billion. Young populations, mobile-first markets, scaling e-commerce and fintech. This is a demographic and digital adoption play spanning Korea and ASEAN.
Consumer Internet (8%): Netflix, Reddit, Spotify. Combined: $2.2 billion. Platform businesses with recurring engagement. Note that Netflix is new (conviction building) while Reddit is being sold (conviction fading). The theme is the same but the individual conviction is diverging.
Vertical SaaS and Fintech (7%): Veeva, Corpay, Flutter, Zscaler, Apollo. Combined: $3.5 billion. High-quality compounders in defensible verticals.
This mapping reveals something the raw position list does not: over 40% of the portfolio is a single thesis on AI infrastructure. That is not diversification. That is a concentrated bet dressed up as multiple positions. Any analyst covering this fund needs to model a scenario where AI infrastructure spending slows. It would hit not one position but seven.
The Exit That Tells You Everything
Sometimes the most instructive data point in a 13F is what is no longer there.
This fund exited Meta after holding it as their number one position for over three years. The reason maps directly from the filing data: Meta announced $100 billion or more in AI capital expenditure for 2026 with unclear returns. The fund had just lived through 2022, where their own overextension cost them $42 billion. They recognized the same pattern in Meta: massive spending outpacing visible monetization. So they left.
That exit teaches you more about the fund’s current psychology than any position they hold. They are believers in AI (40% of portfolio in AI infrastructure). But they are disciplined believers. Spending without visible returns is exactly the behavior that nearly destroyed them four years ago. They will not tolerate it in a portfolio company.
How to Automate This Analysis
The manual version of this work takes a senior analyst several hours per fund per quarter. Pull the 13F from EDGAR, map to previous quarter, calculate changes, classify each position by conviction tier, identify themes, compare exits to theses.
Structured 13F data from providers like Daloopa, FactSet, or direct SEC EDGAR parsing can reduce this to minutes. The conviction tier framework is mechanical: unchanged positions sort to Tier 1, percentage changes determine Tier 2 versus Tier 3, new entries are Tier 4, exits are Tier 5. Theme mapping requires a classification layer, either manual taxonomy or an AI classifier trained on sector data.
The quarterly 13F filing deadline is 45 days after quarter-end. Every fund files in the same window. If you are tracking 20 funds, you need to process 20 13Fs in a compressed timeline, each one requiring the same cross-quarter comparison, tier mapping, and theme extraction. That is either a team of analysts working overtime or a system that does it while they sleep.
The Bottom Line
Position sizes tell you what they own. Conviction tiers tell you what they believe. Theme mapping tells you what they expect. Exits tell you what they fear. One quarterly filing, read properly, reveals an entire investment philosophy.
The fund that reads 20 13Fs in an afternoon while their competitors read two has a structural information advantage. Not from proprietary data. From faster, deeper processing of the same public filings everyone can access.
Frequently Asked Questions
What is a 13F filing and who has to file one?
Form 13F is a quarterly filing required by the SEC for institutional investment managers controlling $100 million or more in qualifying assets. It discloses long equity positions as of the quarter’s end. Filed within 45 days of quarter-end, making Q4 filings due by mid-February.
How do you distinguish profit-taking from loss of conviction?
Look at the remaining position size relative to the portfolio. A 9% trim on a $2.6 billion position still leaves $2.4 billion of conviction. That is valuation discipline. A 63% cut followed by a near-complete exit (like Meta in Q3-Q4 2025) signals a fundamental thesis change. The pattern across quarters, not the single-quarter action, tells the story.
Can 13F filings be misleading?
Yes. 13Fs only show long equity positions at quarter-end. They do not show short positions, options, or any trades that happened mid-quarter. A fund could have bought and sold a position entirely within the quarter with no trace in the 13F. They also show holdings as of a specific date, so positions may have changed by the time the filing becomes public 45 days later.
How do you map portfolio themes from individual positions?
Classify each holding by its primary investment thesis rather than its sector. NVIDIA, Taiwan Semi, and Broadcom are all “semiconductor” companies, but their shared thesis is AI infrastructure. Grouping by thesis rather than sector reveals concentration that standard sector breakdowns miss. In this fund, seven positions across different sub-industries all reduce to one bet: AI infrastructure spending continues.
What tools automate 13F analysis?
EdgarTools (open source, MIT license) provides 13 MCP tools including institutional holdings lookup and fund analysis workflows. Commercial providers like Daloopa and FactSet offer structured 13F data with cross-quarter comparison capabilities. SEC EDGAR provides the raw filings for direct parsing. The conviction tier classification itself is mechanical and straightforward to automate.
Sources: SEC EDGAR 13F filings (Q3-Q4 2025), Tiger Global Management LLC quarterly holdings, Seeking Alpha Q4 2025 portfolio tracking, StockZoa institutional holdings data
Last updated: April 14, 2026
By BetterAI | We build custom AI research infrastructure for European investment firms. See how it works