HALO Logo
HALO
← Back

Trading Strategy

HALO uses an AI-powered automated market maker (AMM) to generate consistent yields through multiple trading strategies.

Overview

Our trading system combines:

  1. Intelligent Market Making: Providing liquidity with optimized spreads on Opinion and Polymarket
  2. Mispricing Detection: AI identifies arbitrage opportunities using ensemble models
  3. Dynamic Spread Adjustment: Real-time optimization using Kelly Criterion and Avellaneda-Stoikov model
  4. Focus Markets: Primarily Opinion and Polymarket, enhancing efficiency in prediction markets

Market Making Strategy

Core Principle

The AMM places simultaneous buy and sell orders around the current market price, capturing the bid-ask spread.

spread = ask_price - bid_price

Spread Optimization

Our AI dynamically adjusts spreads using:

  • Kelly Criterion: Optimal position sizing based on edge and bankroll
  • Avellaneda-Stoikov Model: Real-time spread adjustment based on volatility, liquidity, and market conditions
  • Market volatility analysis
  • Liquidity depth assessment
  • Detected mispricing levels

Spreads are optimized in real-time to maximize fill rates while maintaining profitability, adapting to changing market conditions.

Order Placement

Orders are placed using a sophisticated algorithm that:

  1. Analyzes current market depth
  2. Calculates optimal price levels
  3. Sizes positions based on available capital
  4. Monitors and adjusts in real-time

Mispricing Arbitrage

Detection Algorithm

Our ensemble ML models analyze 45+ features to predict true probability:

  • Historical price data
  • Market sentiment indicators
  • Volume patterns
  • Time to resolution

mispricing = |predicted_probability - market_price|

Execution Criteria

Trades are executed when:

mispricing > 5% AND model_confidence > 60%

This dual-threshold approach ensures we only trade on high-confidence opportunities.

Profit Mechanism

When both legs of a binary market are purchased at prices where P_UP + P_DOWN < 1:

guaranteed_profit = 1 - (P_UP + P_DOWN)

This profit is realized at market settlement, regardless of the outcome.

AI Fair Value Model

Our ensemble model combines:

  • XGBoost: Gradient boosting for feature importance and pattern recognition
  • LSTM: Time series analysis for temporal patterns and sequential dependencies
  • PPO: Reinforcement learning for optimal trading policy
  • LLM: Large language model for sentiment and context analysis
  • Ensemble: Weighted combination for final prediction

Model Performance

The ensemble model analyzes 45 features (25 technical indicators, 4 sentiment signals, 8 funding rates, 8 order book depth metrics) to discover fair value, significantly outperforming naive market prices.

Capital Allocation

Capital is allocated across Opinion and Polymarket using a priority queue system:

  1. High Confidence Arbitrage: Markets with >60% confidence where mispricing is detected
  2. Market Making: Liquid markets with stable spreads and high trading volume
  3. Dynamic Rebalancing: Real-time adjustment based on Kelly Criterion and market conditions

The AI bot maintains balanced inventory and minimizes risks while maximizing spread capture opportunities.

Risk Controls

Each strategy includes built-in risk management:

  • Position Limits: Max 5% of vault per market
  • Stop Losses: 10% loss threshold triggers exit
  • Time Decay: Reduce positions as resolution approaches
  • Correlation Limits: Avoid overexposure to related markets

Performance Metrics

Track strategy performance through:

  • Sharpe Ratio: Risk-adjusted returns
  • Win Rate: Percentage of profitable trades
  • Average Spread Captured: Mean profit per trade
  • Capital Efficiency: Returns per dollar deployed

Next Steps