Agent-Based Simulation as an Alternative to Historical Backtesting

Summary

The document questions whether historical backtesting can adequately test systematic strategies, especially macro-oriented trend strategies. The concern is that the historical record contains only a limited range of events, so it cannot represent every scenario that might occur. The author asks whether existing tools can help test a broader set of conditions or whether an agent-based simulation would need to be built.

The only proposed direction is a speculative suggestion: configure two or more competing agents in a financial setting and explore whether their interaction converges toward a Nash equilibrium. The response provides no model design, implementation details, empirical evidence, or comparison with historical testing. It also cautions that this idea would require substantial domain knowledge. Agent-based simulation is presented as a possibility to investigate, not a demonstrated substitute for historical backtesting or proof that all relevant market scenarios can be covered.

Key ideas

• Historical backtests cannot include events absent from the available record.
• The author considers simulation as a way to explore scenarios beyond observed history.
• One speculative proposal is to model competing agents in a financial context.
• The response offers no evidence that agent interaction will reliably produce useful strategy tests.

Tags

# Fatigue with Historic Backtesting - Alternatives?


# Fatigue with Historic Backtesting - Alternatives?












It seems to me like historic backtesting is the best of bad options out there for me to test my systematic strategies - even ones that are more macro-level trend spotting. I can't test enough scenarios since history is constrained by types of events that have occurred. I want to be robust in my testing and test all possible scenarios that might happen.

Wondering if anyone else feels the same way and whether there are any tools out there or whether I'd have to build my own (ex. Agent-Based Simulations).

## Answer by Adam Conrad (score 2)

https://quant.stackexchange.com/a/75498

Highly speculative and would require a decent degree of domain knowledge, but I would guess that two or more adversarial agents competing against one another in a properly-configured financial context would converge towards a Nash equilibrium relatively quickly.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.