Agent-Based Modeling in AgenticWaves

AgenticWaves implements a sophisticated agent-based modeling (ABM) framework for financial market simulation. This guide explores the ABM components, agent types, and advanced simulation capabilities.

Overview

The ABM framework in AgenticWaves features:

6 distinct behavioral agent types with realistic trading patterns
Heterogeneous agent populations with customizable diversity
Multi-layer network interactions (trading, information, social)
Dynamic wealth distribution with inequality tracking
Crisis-dependent behavior adaptation
Realistic market microstructure with price impact and transaction costs

Agent Types

1. Momentum Traders

Follow price trends and market momentum:

# Momentum traders characteristics
- High trend sensitivity (0.9)
- Moderate risk tolerance (0.7)
- High trading frequency (0.8)
- Short memory length (20 periods)
- Strategy: Buy rising assets, sell falling assets

2. Contrarian Traders

Trade against prevailing market trends:

# Contrarian traders characteristics
- Negative trend sensitivity (-0.7)
- Moderate risk tolerance (0.6)
- Moderate trading frequency (0.5)
- Long memory length (50 periods)
- Strategy: Buy oversold assets, sell overbought assets

3. Fundamentalist Traders

Base decisions on fundamental analysis:

# Fundamentalist traders characteristics
- Low trend sensitivity (0.2)
- Conservative risk tolerance (0.5)
- Low trading frequency (0.3)
- Very long memory (100 periods)
- Strategy: Mean reversion toward fundamental value

4. Noise Traders

Make random or irrational trading decisions:

# Noise traders characteristics
- Very low trend sensitivity (0.1)
- High risk tolerance (0.9)
- Very high trading frequency (0.9)
- Very short memory (5 periods)
- Strategy: Random trading decisions

5. Herding Traders

Follow crowd behavior and social signals:

# Herding traders characteristics
- Moderate trend sensitivity (0.5)
- Moderate risk tolerance (0.6)
- Moderate trading frequency (0.6)
- Short memory length (15 periods)
- Strategy: Follow dominant market sentiment

6. Sophisticated Traders

Use complex multi-factor strategies:

# Sophisticated traders characteristics
- High trend sensitivity (0.6)
- Conservative risk tolerance (0.4)
- Low trading frequency (0.4)
- Very long memory (200 periods)
- Strategy: Multi-factor analysis with risk management

Creating Agent Populations

Basic Population Creation

library(AgenticWaves)

# Create a diverse agent population
agents <- create_enhanced_agent_population(
  n_agents = 500,
  behavioral_heterogeneity = 0.7,
  wealth_distribution = "pareto"
)

# View population summary
print(attr(agents, "type_distribution"))
print(paste("Initial wealth Gini:", round(attr(agents, "wealth_gini"), 3)))

Wealth Distribution Options

# Equal wealth distribution
agents_equal <- create_enhanced_agent_population(
  n_agents = 200,
  wealth_distribution = "equal"
)

# Normal wealth distribution
agents_normal <- create_enhanced_agent_population(
  n_agents = 200,
  wealth_distribution = "normal"
)

# Pareto wealth distribution (realistic inequality)
agents_pareto <- create_enhanced_agent_population(
  n_agents = 200,
  wealth_distribution = "pareto"
)

Behavioral Heterogeneity

The behavioral_heterogeneity parameter controls diversity within agent types:

# Low heterogeneity (similar agents within types)
agents_low_het <- create_enhanced_agent_population(
  n_agents = 300,
  behavioral_heterogeneity = 0.3
)

# High heterogeneity (diverse agents within types)
agents_high_het <- create_enhanced_agent_population(
  n_agents = 300,
  behavioral_heterogeneity = 0.9
)

Multi-Layer Networks

Network Types

AgenticWaves creates three types of agent networks:

# Create multi-layer network
network <- create_dynamic_multilayer_network(
  agents = agents,
  network_types = c("trading", "information", "social"),
  density = 0.1
)

# Access individual networks
trading_network <- network$networks$trading
information_network <- network$networks$information
social_network <- network$networks$social

# View network properties
print(trading_network$metrics)

Trading Network

Connections based on: - Similar agent types - Similar wealth levels
- Similar risk tolerance

Information Network

Connections based on: - Agent sophistication - Memory length - Information processing capabilities

Small-world network with: - Lattice structure with random rewiring - Local clustering with long-range connections - Realistic social interaction patterns

Market Simulation

Basic Simulation

# Load sample data
data <- get_sample_data("global_markets", n_assets = 10, n_periods = 500)

# Run market simulation
sim_results <- simulate_enhanced_market_dynamics(
  agents = agents,
  asset_data = data,
  n_periods = 1000,
  network_effects = TRUE
)

Simulation Components

The simulation includes:

Market Regime Dynamics: Volatility regimes switch based on market conditions
Agent Decision Making: Each agent makes trading decisions based on their type
Network Information Flow: Information spreads through agent networks
Price Impact: Agent trades affect market prices
Transaction Costs: Realistic trading costs reduce profits
Wealth Evolution: Agent wealth changes based on trading performance

Market Regimes

# Market regimes in simulation
# 1 = Low volatility
# 2 = Normal volatility  
# 3 = High volatility/crisis

regime_summary <- table(sim_results$market_regimes)
print(regime_summary)

# Crisis probability and transitions
crisis_periods <- which(sim_results$market_regimes == 3)
print(paste("Crisis periods:", length(crisis_periods)))

Agent Performance Analysis

# Extract agent types
agent_types <- sapply(agents, function(x) x$type)

# Calculate performance by type
performance_by_type <- tapply(sim_results$agent_returns, agent_types, function(x) {
  data.frame(
    mean_return = mean(x),
    volatility = sd(x),
    sharpe_ratio = mean(x) / sd(x),
    max_return = max(x),
    min_return = min(x)
  )
})

# View performance summary
print(performance_by_type)

Wealth Inequality Evolution

# Track wealth inequality over time
wealth_gini_evolution <- sim_results$wealth_evolution_gini

# Plot wealth inequality
plot(wealth_gini_evolution, 
     type = "l", 
     main = "Wealth Inequality Evolution",
     xlab = "Time Period", 
     ylab = "Gini Coefficient")

# Final wealth distribution
final_wealth <- sim_results$agent_wealth[nrow(sim_results$agent_wealth), ]
final_gini <- calculate_gini_coefficient(final_wealth)

print(paste("Final wealth Gini:", round(final_gini, 3)))

Advanced Agent Behaviors

Crisis Adaptation

Agents adapt their behavior during crisis periods:

# During crisis (regime 3):
# - Risk tolerance decreases
# - Trading frequency may increase (flight to quality)
# - Herding behavior intensifies
# - Fundamental analysis becomes less reliable

Agents learn from connected neighbors:

# Information sharing through networks:
# - Sophisticated agents share more information
# - Information quality depends on agent type
# - Social influence affects trading decisions
# - Network structure impacts information flow

Memory and Learning

Each agent maintains memory of past market conditions:

# Memory effects:
# - Recent events weighted more heavily
# - Memory length varies by agent type
# - Learning rate affects adaptation speed
# - Performance feedback influences future decisions

Customizing Agent Behavior

Creating Custom Agent Types

While AgenticWaves provides 6 built-in agent types, you can customize behavior:

# Modify agent parameters after creation
custom_agents <- create_enhanced_agent_population(n_agents = 100)

# Customize specific agents
for(i in 1:length(custom_agents)) {
  if(custom_agents[[i]]$type == "momentum") {
    # Make momentum traders more aggressive
    custom_agents[[i]]$risk_tolerance <- custom_agents[[i]]$risk_tolerance * 1.2
    custom_agents[[i]]$trading_frequency <- custom_agents[[i]]$trading_frequency * 1.1
  }
}

Parameter Sensitivity Analysis

Test how different parameters affect outcomes:

# Test different behavioral heterogeneity levels
heterogeneity_levels <- c(0.3, 0.5, 0.7, 0.9)
results_by_heterogeneity <- list()

for(het in heterogeneity_levels) {
  agents_test <- create_enhanced_agent_population(
    n_agents = 200,
    behavioral_heterogeneity = het
  )
  
  sim_test <- simulate_enhanced_market_dynamics(
    agents = agents_test,
    asset_data = data,
    n_periods = 500
  )
  
  results_by_heterogeneity[[as.character(het)]] <- sim_test$final_wealth_gini
}

print(results_by_heterogeneity)

Visualization of Agent Networks

Agent Network by Type

# Create agent network visualization
network_plot <- plot_agent_type_network(
  agents = agents,
  network_obj = network$networks$trading$graph,
  layout = "stress"
)

print(network_plot)

Wealth Distribution Plots

# Plot wealth distribution by agent type
wealth_by_type <- data.frame(
  agent_type = sapply(agents, function(x) x$type),
  wealth = sapply(agents, function(x) x$current_wealth)
)

# Wealth distribution plot
library(ggplot2)
ggplot(wealth_by_type, aes(x = agent_type, y = wealth, fill = agent_type)) +
  geom_boxplot() +
  scale_y_log10() +
  theme_minimal() +
  labs(title = "Wealth Distribution by Agent Type",
       x = "Agent Type", y = "Wealth (log scale)")

Performance Optimization

Simulation Speed

For large simulations, consider:

# Optimize for speed
# - Reduce number of agents for initial testing
# - Use shorter simulation periods
# - Disable network effects for simple simulations
# - Use parallel processing where available

# Fast simulation for testing
quick_sim <- simulate_enhanced_market_dynamics(
  agents = agents[1:100],  # Subset of agents
  asset_data = data,
  n_periods = 250,         # Shorter period
  network_effects = FALSE  # Disable networks
)

Memory Management

# For large simulations:
# - Monitor memory usage
# - Clear intermediate results
# - Use appropriate data types
# - Consider chunked processing for very long simulations

# Monitor memory (if pryr package available)
if(requireNamespace("pryr", quietly = TRUE)) {
  print(pryr::mem_used())
}

Research Applications

The ABM framework supports various research questions:

Market Efficiency Studies

Impact of different trader types on price discovery
Role of information networks in market efficiency
Behavioral biases and market anomalies

Systemic Risk Analysis

Contagion mechanisms through agent networks
Impact of wealth inequality on market stability
Crisis propagation and amplification

Policy Analysis

Effects of transaction taxes on different agent types
Impact of market structure changes
Regulatory interventions and market outcomes

Behavioral Finance

Herding behavior and bubble formation
Momentum vs. contrarian strategies
Learning and adaptation in financial markets

Next Steps

Explore Network Analysis for spillover detection
Read Visualization Guide for advanced plotting
Try Advanced Analysis for complex workflows
Check Custom Data tutorial for your own datasets

References

Axtell, R. L., & Farmer, J. D. (2025). Agent-based modeling in economics and finance: Past, present, and future.
Hommes, C. (2006). Heterogeneous agent models in economics and finance.
LeBaron, B. (2006). Agent-based computational finance.
Farmer, J. D., & Foley, D. (2009). The economy needs agent-based modelling.

Agent-Based Modeling in AgenticWaves