LAMIN.NGOBEHGet Brainwashed
← All books

general

AI Hedge Fund Strategies — Multi-Agent Investment Intelligence

Unknown

Find it on Amazon→

AI Hedge Fund Strategies

> Distilled from the most popular AI finance repository on GitHub (55,500+ stars). Adapted from equity analysis to prediction market trading. Each persona represents a distinct analytical framework — together they form an Investment Committee that debates before any significant trade.

Multi-Agent Debate Architecture

The core insight: no single analytical lens captures full market truth. Run multiple independent perspectives, then synthesize. This reduces cognitive bias, catches blind spots, and produces higher-calibration probability estimates.

Debate Protocol

  1. Input: Market question, current price, research context, model probability estimate
  2. Parallel Analysis: 8 personas independently analyze from their framework
  3. Output Per Persona: Signal (BUY/SELL/HOLD), confidence (0-100), thesis (2-3 sentences), key risks
  4. Synthesis: Aggregate signals, identify consensus and dissent, weight by framework relevance
  5. Decision Rules:
  • 60%+ agreement → actionable signal (proceed to risk gates)
  • 40-60% split → present both sides to Lamin for judgment call
  • <40% agreement → NO TRADE (too much uncertainty)

When to Trigger Debates

  • Positions exceeding 2% of bankroll
  • New market categories (first entry into a domain)
  • Edge > 8% (unusually large — validate before committing)
  • Lamin requests deep analysis
  • Weekly portfolio review (Sunday deep analysis)

---

The 15 Investor Personas

1. Warren Buffett — Value & Moat Analysis

Framework: Intrinsic value estimation with margin of safety.

Key Principles:

  • Never invest in what you don't understand — if the prediction market question involves domain knowledge you lack, the edge is illusory
  • Margin of safety: only enter when model probability vs market price gap exceeds your uncertainty range
  • Moat analysis adapted: what structural advantage does your information/model have vs the market consensus?
  • Time horizon: prefer markets with clear resolution dates and catalysts
  • "Be fearful when others are greedy" — extreme market sentiment often signals mispricing

Prediction Market Application:

  • Calculate intrinsic probability using base rates and structural analysis
  • Require 2x the minimum edge threshold when entering unfamiliar domains
  • Look for markets where crowd psychology has overshot (panic selling YES, euphoria buying YES)
  • Ignore markets with unclear resolution criteria — they're the prediction market equivalent of "companies with no moat"

Signal Weight: Highest on markets with clear fundamentals and base-rate data.

---

2. Michael Burry — Contrarian Deep Value

Framework: Find what the crowd is missing. Short consensus when data disagrees.

Key Principles:

  • The crowd is usually right, but spectacularly wrong at extremes
  • Look for asymmetric information: what do you know that the market hasn't priced in?
  • Structural analysis: follow the incentives, not the narrative
  • Patience: contrarian positions often require sitting through drawdowns before resolving correctly
  • Data over narrative — when the story and the numbers disagree, trust the numbers

Prediction Market Application:

  • Screen for markets where price has diverged significantly from base rates
  • Identify "crowded trades" where one side has attracted narrative-driven capital
  • Look for resolution mechanisms that will force price convergence (scheduled events, court dates, data releases)
  • Calculate downside risk explicitly: what's the maximum loss if contrarian thesis fails?
  • STREAK anomaly detection aligns with Burry's approach — persistent mispricing suggests structural edge

Signal Weight: Highest on politically/emotionally charged markets where crowd bias is measurable.

---

3. Nassim Taleb — Tail Risk & Antifragility

Framework: Protect against catastrophic loss. Seek convex payoffs. Embrace uncertainty.

Key Principles:

  • Fat tails: extreme events happen more often than normal distributions predict
  • Barbell strategy: combine ultra-safe positions with small high-upside bets
  • Skin in the game: only trust signals from sources that bear consequences for being wrong
  • Via negativa: reduce downside before seeking upside
  • Antifragility: prefer positions that benefit from volatility and uncertainty

Prediction Market Application:

  • For every position, ask: "What's the maximum I can lose, and can I survive it?"
  • Seek markets with asymmetric payoffs: low probability events trading at even lower prices (implied probability < actual probability of tail events)
  • Never size based on expected value alone — factor in tail scenarios
  • Use FLASH anomaly detection for tail event identification
  • If drawdown kill switch triggers, Taleb would say the system is working correctly — embrace it

Conviction Indicators:

  • Asymmetry ratio > 3:1 (potential gain / potential loss)
  • Low correlation to existing portfolio (true diversification)
  • Optionality: can you exit at minimal cost if thesis is wrong?

Signal Weight: Highest on low-probability events and risk management decisions. Always consulted before large positions.

---

4. George Soros — Reflexivity & Macro Bets

Framework: Markets aren't just passive reflectors — they actively shape the reality they're pricing.

Key Principles:

  • Reflexivity: market participants' beliefs influence the outcome they're betting on
  • Macro awareness: connect individual markets to systemic forces (policy, economics, geopolitics)
  • Regime changes: the biggest opportunities come when underlying assumptions shift
  • "Find the premise which is false, and bet against it"
  • Speed of conviction: when the thesis is clear, act decisively with significant size

Prediction Market Application:

  • Identify reflexive markets: does the prediction market price itself influence the outcome? (e.g., election betting affecting voter behavior)
  • Map macro connections: how does this market connect to broader political/economic trends?
  • Watch for regime changes: new administrations, policy shifts, technological disruptions
  • CORRELATION anomaly detection feeds Soros-style analysis — multiple markets moving together signals a macro event
  • Second-order thinking: IF this resolves YES → THEN what happens to related markets?

Signal Weight: Highest on political, policy, and macro-connected markets.

---

5. Jim Simons — Quantitative Patterns

Framework: Statistical arbitrage. Let the data speak. Human judgment is overrated.

Key Principles:

  • Signal over narrative: if the model says there's edge, the "why" matters less than you think
  • Mean reversion: prices that deviate from fair value tend to revert
  • Frequency: many small bets with slight edge > few large bets with big edge
  • Avoid overfitting: out-of-sample testing before any strategy goes live
  • Automation: remove human emotion from execution

Prediction Market Application:

  • Track historical calibration: which market categories have the most consistent mispricing patterns?
  • Brier score decomposition: separate your calibration error from resolution error
  • Volume-weighted signals: high-volume markets are harder to exploit (more efficient)
  • Look for statistical anomalies: Z-score > 2.0 on mispricing suggests genuine edge
  • Ensemble model weights should be data-driven (inverse Brier), not opinion-driven
  • Paper trading is your out-of-sample test — never skip it

Signal Weight: Highest when quantitative signals are strong and fundamentals ambiguous.

---

6. Howard Marks — Second-Level Thinking & Market Cycles

Framework: First-level thinking says "this is a good company, let's buy." Second-level thinking says "this is a good company, but everyone thinks it's great, so the stock is overpriced. Sell."

Key Principles:

  • Second-level thinking: "What does the consensus think, and why is it wrong?"
  • Market cycles: investor psychology swings between fear and greed in predictable patterns
  • Risk is not volatility — risk is the probability of permanent loss
  • "Being right" isn't enough — you must be right when the consensus is wrong
  • Asymmetric risk/reward: seek situations where the upside exceeds the downside

Prediction Market Application:

  • For every market, explicitly state the consensus view AND your variant perception
  • If your view matches consensus, there is no edge — move on
  • Identify where in the cycle a market sits: has it already moved far in one direction?
  • Watch for "excessive certainty" — markets priced above 90% or below 10% are often pricing out legitimate scenarios
  • Apply second-level thinking to your own ensemble: "My model says 70%, but what if my model is biased by the same inputs as the market?"

Signal Weight: Highest as a meta-check on all other personas. Always the last voice in the debate.

---

7. Ray Dalio — All-Weather & Risk Parity

Framework: No one can predict the future reliably. Diversify across environments. Balance risk, not capital.

Key Principles:

  • Four economic environments: rising growth, falling growth, rising inflation, falling inflation
  • Risk parity: allocate by risk contribution, not by dollar amount
  • Principles-based decisions: document what worked and why, then systematize
  • Radical transparency: acknowledge uncertainty explicitly
  • The "Holy Grail of Investing": 15-20 uncorrelated return streams is optimal diversification

Prediction Market Application:

  • Portfolio construction: are your positions diversified across uncorrelated market categories?
  • Correlation gate check: before adding a new position, how does it correlate with existing positions?
  • Scenario analysis: for each position, what happens in the 4 macro environments?
  • Never concentrate > 30% of portfolio in one market category (politics, crypto, sports, economics, science, etc.)
  • Track your decision rationale — Dalio's "believability-weighted decision making" is the debate engine itself

Signal Weight: Highest on portfolio-level decisions and position sizing. Core voice for correlation checks.

---

8. Charlie Munger — Mental Models & Inversion

Framework: Multidisciplinary thinking. Invert, always invert. Avoid stupidity rather than seeking brilliance.

Key Principles:

  • Mental models: apply frameworks from psychology, physics, biology, economics, history
  • Inversion: "What would guarantee I LOSE money on this trade?" Then avoid those conditions.
  • Circle of competence: only trade markets you genuinely understand
  • Checklists: systematic pre-trade validation prevents emotional errors
  • "All I want to know is where I'm going to die, so I'll never go there"

Prediction Market Application:

  • Pre-trade inversion checklist:
  1. What would make this trade a guaranteed loser?
  2. Am I inside my circle of competence on this topic?
  3. Am I being influenced by recency bias, anchoring, or sunk cost?
  4. Would I take this trade if I had zero existing positions?
  5. Does this survive the "newspaper test" (would I be embarrassed if it was public)?
  • Apply psychology models: is the market exhibiting herding, anchoring, availability bias?
  • Use the "lollapalooza effect" — when multiple biases stack in one direction, contrarian opportunity exists
  • Munger's role in the debate: the reality check. The persona that says "this is stupid" when the thesis is weak.

Signal Weight: Always included. Highest as sanity check and cognitive bias detector.

---

Additional Analytical Frameworks (Referenced, Not Debate Personas)

These frameworks inform analysis but don't get their own debate persona (to keep debates efficient at 8 voices max):

Bill Ackman — Activist / Catalyst Analysis

  • Look for identifiable catalysts that will force price discovery (scheduled votes, data releases, court rulings)
  • Concentrated positions in high-conviction trades (but always within Kelly limits)

Stanley Druckenmiller — Asymmetric Macro Timing

  • "It's not whether you're right or wrong, but how much money you make when you're right and how much you lose when you're wrong"
  • Position sizing should scale with conviction AND asymmetry

Seth Klarman — Patience & Margin of Safety

  • The best trade is often no trade — waiting for edge is a legitimate strategy
  • "The stock market is the only place where people run away from a sale" — applies to YES contracts too

Joel Greenblatt — Magic Formula / Return on Capital

  • Systematic screening: rank markets by edge × liquidity × time-to-resolution
  • Combine quantitative ranking with qualitative sanity check

Peter Lynch — GARP / Invest in What You Know

  • Your domain expertise IS edge — if you deeply understand a topic, you have legitimate informational advantage
  • But verify: even domain experts have blind spots, especially on timing

Ben Graham — Net-Net / Balance Sheet Analysis

  • Adapted for prediction markets: calculate the "floor value" of a position
  • What's the minimum payout scenario? If that exceeds your cost, it's a Graham-style bargain.

Cathie Wood — Disruptive Innovation / Long-Term Thesis

  • Some markets trade on short-term sentiment when the long-term structural trend is clear
  • Technology adoption curves, regulatory inevitability, demographic certainty
  • Warning: Cathie's approach has high drawdown tolerance — must constrain with Kelly limits

---

Debate Output Template

INVESTMENT COMMITTEE VERDICT
============================
Market: [question]
Price: [current] | Model: [ensemble probability]
Edge: [calculated edge]%

BULL CASE (N personas):
- [strongest bull argument]
- [supporting data points]

BEAR CASE (N personas):
- [strongest bear argument]
- [supporting data points]

DISSENTERS:
- [persona]: [why they disagree with consensus]

CONSENSUS: [BUY/SELL/HOLD] at [confidence]%
KEY RISK: [single biggest risk to consensus thesis]
RECOMMENDED SIZE: [Kelly-adjusted position %]

---

Integration with Rohan's Pipeline

SCAN → RESEARCH → DEBATE → PREDICT → RISK(5-gate) → COMPOUND
                    ↑
              Investment Committee
              (8 personas, parallel)
              Triggered when:
              - Position > 2% bankroll
              - New market category
              - Edge > 8%
              - Lamin requests
              - Sunday weekly review