P2A-C4 · Portfolio Construction¶
Core One-Liner
Position sizing matters more than stock picking — 90% of retail investors lose on sizing, not picking.
Universal Investment Model — Works for Any Industry
P2A-C4 (Part 2.A Chapter 4). After this chapter, you can properly size a 5-10 stock portfolio and stop going all-in on a single stock.
1. The Problem: 5 Stocks at 20% Each — Big Mistake¶
Common retail portfolios: - 5 stocks at 20% each (equal sizing — no distinction between high/low conviction) - Or 1 stock at 50%+ (all-in, extremely concentrated) - Or 30+ stocks at 3% each (over-diversified, no different from an ETF)
All 3 are wrong: - Equal sizing doesn't leverage conviction (your high conviction stocks should be larger) - All-in carries too much risk (one stock goes bad, entire portfolio suffers) - Over-diversified exceeds marginal benefit (>15 stocks, risk reduction plateaus)
→ Sizing Decision: Position size ∝ conviction × payoff ÷ risk.
2. Solution: 3 Sizing Frameworks¶
| Framework | How to Size | Best For |
|---|---|---|
| Kelly Criterion | size = (edge × payoff) / risk | Math-oriented, impractical (too aggressive) |
| Concentration (Buffett) | Top 5 make up 80%+ | High conviction, long-term hold |
| Risk Parity (Bridgewater) | Each stock contributes equal risk | Sector balancing |
Practical Recommendation: Modified Buffett Concentration: - Top 5 holdings: 60-80% - Single stock max: 15-20% - Starter position: 1-3% - Core position: 5-10% - Max conviction: 10-15%
3. How It Works: 3 Frameworks in Detail + AI Portfolio Application¶
3.1 Kelly Criterion (Mathematically Optimal)¶
Formula:
f* = (p × b - q) / b
f* = optimal position (% of capital)
p = probability of winning
b = payoff on a win (win $1 + win ratio)
q = probability of losing = 1 - p
Example: You think NVDA over 1 year: - 60% probability +30% - 40% probability -20%
Kelly: f = (0.6 × 1.5 - 0.4) / 1.5 = 0.33 = 33% position*
Why It's Not Practical: - Too aggressive (33% in a single stock is not risk-tolerant) - p / b estimates are highly uncertain (how do you know 60% / 30%?) - Practical recommendation: use Half Kelly (16-17%) or Quarter Kelly (8%)
3.2 Concentration (Buffett)¶
Buffett Portfolio (Berkshire's public 13F): - AAPL ~40% (before trimming) - BAC ~10% - KO ~8% - AXP ~7% - Others ~30% diversified
Top 5 make up 75-80%.
Munger's Quote (1990s speech):
"Diversification is for people who don't know what they are doing."
Buffett Framework: - High conviction (passes all 5 steps, P3-C2) → core 10-15% - Medium conviction → 5-10% - Speculative → < 3%
Advantages: Leverages conviction, long-term outperformance Risk: If a single stock fails, the hit is severe (e.g., AAPL 2024 -20% → Berkshire -8%)
3.3 Risk Parity (Bridgewater)¶
Dalio Framework: - Balance by risk contribution, not dollar size - Each position contributes equal portfolio volatility - Use leverage to balance low-vol assets (bonds) with high-vol (stocks)
Practical Calculation Requires: - Standard deviation of each stock - Correlation matrix - Complex, difficult for retail investors
Simplified Version: Give high-vol AI stocks (NVDA / CRWV) small size, give low-vol (MSFT / GOOGL) large size.
4. AI Portfolio in Practice — 5-10 Stock Sample¶
4.1 5-Stock Conservative (60% AI, 40% Diversification)¶
Core (60%):
- MSFT 15% (low vol AI + cloud)
- GOOGL 12% (low vol AI + Search)
- NVDA 10% (high vol AI core, smaller due to risk)
- AAPL 10% (Buffett-style holding)
- AVGO 8% (AI infrastructure + diversification)
Diversification (40%):
- BRK.B 10% (Buffett, cash hoard exposure)
- KO 5% (defensive)
- XLE 5% (energy hedge)
- BND 10% (bond)
- Cash 10%
4.2 10-Stock Aggressive (80% AI, 20% Diversification)¶
AI Core (60%):
- NVDA 12% / MSFT 12% / GOOGL 10% / AVGO 8% / AAPL 8% / TSM 5% / META 5%
AI Ramp (20%):
- ANET 5% / CRWV 3% / CEG 5% / VRT 4% / COHR 3%
Diversification (20%):
- BRK.B 8% / BND 7% / Cash 5%
4.3 Single Stock Max + Sector Cap¶
Buffett-Style Limits: - Single stock max: 15-20% - Same sector (e.g., AI infra) max: 50% - Same supply chain (e.g., NVDA + SK Hynix + COHR + ASML all affected by NVDA capex) max: 40%
→ Avoid single point of failure hurting the entire portfolio.
5. Sizing vs. Conviction Relationship¶
| Conviction (P3 Tool Score) | Sizing |
|---|---|
| 5/5 (Buffett all pass + 3 PM bull + Anti all ok) | 10-15% |
| ⅘ | 5-10% |
| ⅗ | 3-5% |
| ⅖ | 1-3% starter |
| ⅕ | Don't invest / very small trade |
Example: - MSFT (P3 score 4.5/5) → 10-12% - NVDA (P3 score 3.5/5, Buffett fails) → 5-8% - OpenAI (private, can't invest publicly) → N/A - CEG (P3 score ⅘) → 5-8%
6. Rebalancing Rhythm (Covered in P3-C1 Step 3)¶
- Don't go all-in at once — starter → ramp → core over 6-12 weeks
- Quarterly rebalance — price moves cause sizing drift, adjust quarterly
- Thesis invalidated → trim 50% → clear within 1-2 months (P3-C1 Step 5)
7. Try It: Size a 5-10 Stock Portfolio¶
Task (~1 hr):
| Stock | P3 Score | My Conviction | Sizing |
|---|---|---|---|
| ___ | ?/5 | high/med/low | __% |
| ___ | ?/5 | __% | |
| ___ | ?/5 | __% | |
| ___ | ?/5 | __% | |
| ___ | ?/5 | __% | |
| Cash | __% |
Self-Check (3 items met → proceed to P2A-C5):
- Your top 5 total size 60-80% (not equal, leveraging conviction)
- Single stock max ≤ 15-20% (prevent single point of failure)
- Cash + defensive ≥ 15-20% (keep dry powder)
8. What's Next¶
Sizing is a technique. But your biggest opponent isn't the market — it's you. Behavioral biases cause you to over-size / panic sell / FOMO buy.
→ P2A-C5 · Behavioral Finance — The 6 biases you're most likely to fall into.
9. Deep Dive (optional): Half Kelly / Max Drawdown / Munger's Mental Model¶
Click to see 3 deep dives on sizing
Half Kelly Practical Calculation: - Full Kelly assumes you know p / b with 100% certainty - In reality, p / b are estimates with ±20% error - Half Kelly = half the size, providing robustness - Even if Kelly shows 33%, using 16% is safer
Max Drawdown Thinking: - What's the worst 1-year decline in your portfolio's history? - Can you psychologically handle -30%? -50%? -70%? - If you can't handle -50%, don't go 100% AI concentrated - Adding 20% bonds / cash → -50% becomes -35%
Munger's Mental Model on Recovery: - A position down -50% needs +100% to break even - A position down -75% needs +300% to break even - Downside asymmetry — large drawdowns are far more lethal than small ones - → When sizing, preemptively avoid catastrophic loss