AI-Powered Synthesis of Austrian and Chicago ideas
Like many free-market economists, I began my early studies deeply influenced by the Austrian School, but over time I shifted toward the Chicago School, particularly when it comes to microeconomic theory.
These days, I consider myself primarily aligned with Chicago Price Theory on microeconomic matters, while my views on macroeconomics remain firmly in the Monetarist camp as regular followers of this blog should well-know.
That said, I haven’t completely abandoned the Austrian School. I still believe that Austrian economics offers essential insights into how real-world markets function.
Traditional neoclassical theory—especially Chicago Price Theory—excels at explaining equilibrium outcomes but does little to describe how markets actually work in the real world, where information is incomplete and dispersed. This is where the Austrian School, particularly through the works of F. A. Hayek and Israel Kirzner, provides valuable perspective by framing the market as a discovery process.
In this article, I introduce a new approach that I call Agent-Based Experimental Economics (ABEE).
ABEE blends the insights of both neoclassical economics and the Austrian School with modern artificial intelligence.
Using AI agents that learn and adapt over time, we can simulate market dynamics and observe how these agents interact in ways that are far more realistic than traditional economic models. ABEE captures not just the end state of markets, but the process of getting there—how agents discover information, adapt their strategies, and refine their behaviours.
The Experiment: AI Monopolist Learning in Action
In this particular experiment, we simulated a monopolist using reinforcement learning.
Reinforcement Learning (RL) is a type of machine learning where an agent learns by interacting with an environment. The agent takes actions, receives rewards or penalties, and aims to maximize its long-term reward by improving its decisions over time
The monopolist starts with no knowledge of the market and learns over time how to set prices through trial and error.
This simulation was coded in Python, with a little help from ChatGPT, which allowed us to create an adaptive AI that evolves its strategy based on profit feedback. See the code here.
Here’s how the experiment is structured:
Clik here to view.
Theoretical Expectations vs. Learning Process
Traditional economic theory provides clear predictions for a profit-maximising monopolist: set marginal revenue equal to marginal cost.
Given the demand curve and marginal cost MC=20 the monopolist should charge P*=60 and sell Q*=20 units.
In this experiment, however, the AI monopolist does not have access to this information.
Instead, it learns through exploration—setting different prices and observing the profit outcomes. Over time, the AI agent gradually “discovers” the optimal pricing strategy.
Results: Learning Toward Optimal Pricing
The results, visualised in the graph below, show how the AI monopolist learns to behave optimally (or very close to optimally) over time.
Clik here to view.
Initially, the agent’s pricing decisions are random and far from optimal, but as it gains experience, it converges toward the theoretical profit-maximising price and quantity.
- Price Evolution: The blue line in the graph represents the AI agent’s price decisions over time. Early on, the prices fluctuate as the monopolist explores different pricing strategies. However, as the episodes progress, the prices stabilise and converge toward the theoretical price of P=60 (shown by the red-dashed line).
- Quantity Evolution: The corresponding quantities sold also fluctuate early on as prices vary, but the AI agent gradually learns to settle around the optimal quantity of Q=20Q = 20Q=20.
It is also notable that once the AI agent reaches the optimal price and quantity, it doesn’t change the price significantly.
These results confirm that an AI agent, even starting with no knowledge of the market, can learn to mimic the behaviour predicted by traditional neoclassical economics.
Through trial and error, the AI monopolist discovers the price that maximises profit, demonstrating that markets can converge to the theoretically correct equilibrium even when participants begin with limited information.
Agent-Based Experimental Economics: A New Lens for Economic Research
The concept of Agent-Based Experimental Economics (ABEE) represents a new tool for exploring economic behaviour.
Traditional economic models often rely on simplifying assumptions, like the “representative agent,” to describe equilibrium outcomes. But ABEE allows us to simulate dynamic markets populated by AI agents that learn and adapt in real time, reflecting the diversity of decision-making we see in the real world.
In this experiment, we only simulated a single monopolist and fixed consumer behaviour. However, the potential of ABEE goes far beyond this simple model.
ABEE allows us to simulate large-scale economies with diverse agents, each learning and interacting in complex ways. This opens the door to a whole new range of experiments, from modelling market competition to testing different monetary policy rules.
The Market as a Process: Connecting Neoclassical and Austrian Insights
One of the things that to me makes ABEE particularly exciting is how it bridges the gap between neoclassical economics and Austrian insights.
While neoclassical theory is excellent at predicting equilibrium outcomes, it often overlooks how markets reach these outcomes. The Austrian School, particularly the works of Hayek and Kirzner, focuses on the market process—the idea that markets are constantly evolving as individuals learn and adapt.
In our simulation, the AI monopolist reflects this discovery process. The monopolist doesn’t start with perfect information or rational expectations but learns the optimal strategy through repeated interactions with the market.
This mirrors Hayek’s view of the market as a discovery mechanism, where decentralised information is gradually revealed through the actions of market participants.
Furthermore, the simulation demonstrates Israel Kirzner’s concept of entrepreneurial alertness—the idea that markets correct themselves through the discovery of previously unexploited opportunities. The AI monopolist gradually “discovers” the optimal price, much like how entrepreneurs discover profitable niches in real-world markets.
This dynamic approach also echoes Ludwig von Mises’ idea of the “Evenly Rotating Economy,” where equilibrium is a theoretical construct and real markets are always in flux. ABEE captures this dynamism by simulating markets as evolving systems rather than static, equilibrium-based models.
Applications of ABEE: From Policy to Strategy
The potential applications of ABEE extend well beyond academic research. Governments, businesses, and policymakers can use ABEE to simulate real-world scenarios, providing valuable insights into how markets react to policy changes, business strategies, and economic shocks.
- Monetary Policy: Central banks could use ABEE to simulate the impact of different monetary policy rules on diverse agents.
- Labour Market Studies: Policymakers can use ABEE to model the effects of minimum wage increases or changes in labour laws, capturing the varied responses of workers with different skills and backgrounds.
- Business Strategy: Companies can use ABEE to test pricing strategies, product launches, or supply chain decisions in a simulated environment, observing how AI agents with diverse preferences react to these changes.
- Income Distribution: Governments can use ABEE to simulate the effects of tax reforms or welfare programs, providing deeper insights into how different groups are affected by these policies.
Conclusion: ABEE as a Transformative Tool for Economics
Agent-Based Experimental Economics (ABEE) offers a powerful new framework for understanding market behaviour. By simulating dynamic markets where agents learn and adapt, ABEE allows us to explore the market process in ways that traditional static models cannot.
In this simulation, the AI monopolist learned to set prices that mirrored the predictions of neoclassical theory. But beyond simply confirming existing models, ABEE also provides new insights into how markets evolve over time, much in line with Austrian economics.
The results—visualised in our graphs—demonstrate that, even in the absence of perfect information, markets have the potential to spontaneously organize towards stable outcomes.
This aligns with Hayek’s and Kirzner’s views of markets as discovery processes, where entrepreneurs and firms learn by interacting with the market.
Looking forward, the potential of Agent-Based Economic Experiments (ABEE) to revolutionize economic research, policy-making, and business strategy is truly exciting.
The future of economic experimentation lies in these dynamic simulations, and with the help of AI, we are only scratching the surface of what’s possible.
These advanced tools promise to provide unprecedented insights into complex economic systems, allowing for more accurate predictions and better-informed decision-making across various sectors.
* Needless to say, my thinking these topics is greatly inspired by the great Vernon Smith – the father of traditional Experimental Economics.
———————
If you want to know more about my work on AI and data, then have a look at the website of PAICE — the AI and data consultancy I have co-founded.
Clik here to view.
