Agent-Based Financial Market Simulation

Stochastic vs Deterministic Trading Strategies

This project implements an agent-based simulation of a financial market to study the evolutionary dynamics between deterministic and stochastic trading strategies.

Project Structure

├── market.py              # Market mechanics and price formation
├── traders.py             # Trader classes and strategies
├── simulation.py          # Simulation engine and analysis
├── run_experiments.py     # Experiment runner
├── results/              
└── README.md             

Components

1. Market Environment (market.py)

The market simulation includes:

• 4 Stocks with different characteristics:

  • Stock A: High volatility, trending
  • Stock B: Mean-reverting, low volatility
  • Stock C: Random walk
  • Stock D: Cyclical patterns

• Price Formation: Prices are determined by:

  • Supply/demand imbalance from trader orders
  • Random noise component
  • Stock-specific characteristics (trend, mean reversion, cycles)
  • Price impact coefficient

2. Trading Strategies (traders.py)

With the exception of Adaptive Stochastic Traders (since they are already "learning" from the market), after 5 consecutive losses, traders evaluate all strategies (with the exception of Adaptive Stochastic) using a Sharpelike score and switch if another strategy outperforms the current one by a fixed threshold.

Deterministic Strategies

1. Momentum Trader: Buys rising stocks, sells falling ones

   • Uses lookback period to calculate returns
   • Trades when momentum exceeds threshold

2. Mean Reversion Trader: Assumes prices revert to historical mean

   • Calculates z-scores of current prices
   • Buys when oversold, sells when overbought

3. Value Trader: Trades based on fundamental value estimates

   • Maintains internal estimates of stock values
   • Buys undervalued, sells overvalued stocks

Stochastic Strategies

1. Fixed Stochastic Trader: Uses fixed probability matrix

   • Randomly selects from deterministic strategies
   • Probabilities remain constant throughout simulation

2. Adaptive Stochastic Trader: Dynamically adjusts strategy weights

   • Tracks performance of each strategy
   • Updates probabilities based on recent returns
   • Uses softmax transformation for weight updates

3. Simulation Engine (simulation.py)

• MarketSimulation: Core simulation class

  • Manages traders and market interactions
  • Tracks performance metrics
  • Checks for convergence to equilibrium

• ExperimentRunner: Manages multiple experiments

  • Tests different trader distributions
  • Saves results and generates plots
  • Analyzes equilibrium patterns

Running the Experiments

Requirements

conda install numpy pandas matplotlib seaborn (or pip if conda is not available)

Basic Usage

python run_experiments.py

This will:

1. Run multiple experiments with different trader distributions
2. Generate plots for each experiment
3. Save detailed results to JSON files
4. Create a summary CSV with key metrics
5. Print equilibrium analysis

Custom Experiments

from simulation import MarketSimulation

# Define custom configuration
config = {
    'initial_cash': 10000,
    'risk_tolerance': 0.5
}

# Define trader distribution
distribution = {
    'momentum': 20,
    'mean_reversion': 20,
    'value': 20,
    'fixed_stochastic': 20,
    'adaptive_stochastic': 20
}

# Run simulation
sim = MarketSimulation(config)
results = sim.run_simulation(distribution, max_periods=1000)

# Plot results
sim.plot_results(results, 'my_results.png')

Key Assumptions

1. Market Structure:

   • Perfect liquidity (can always trade at market price)
   • No transaction costs
   • Synchronous trading (all agents trade simultaneously)
   • Public information (all see same price history)

2. Trader Behavior:

   • Traders are price takers (no market manipulation)
   • Fixed total capital in the system
   • No leverage or short selling
   • Risk tolerance affects position sizing

3. Price Formation:

   • Prices respond to supply/demand imbalance
   • Each stock has inherent characteristics
   • Random noise represents external factors

Output Analysis

The simulation produces several outputs:

1. Convergence Analysis: Whether strategy proportions stabilize
2. Performance Metrics: Average returns and Sharpe ratios by strategy
3. Evolution Plots: How strategies and prices evolve over time
4. Distribution Analysis: Final (equilibrium, if it is) proportions