Forecasting from Counterfactual Simulator Rollouts: A Sim2Real Evaluation
Deploying a new decision policy creates a cold-start problem for prediction models whose targets depend on the policy's actions: historical observations reflect earlier policies, while real observations under the new policy are not yet available. Simulation offers a way to address this gap by rolling out the target policy across counterfactual scenarios and using the resulting
Reading in Plain English mode — toggle below to switch
Deploying a new decision policy creates a cold-start problem for prediction models whose targets depend on the policy's actions: historical observations reflect earlier policies, while real observations under the new policy are not yet available. Simulation offers a way to address this gap by rolling out the target policy across counterfactual scenarios and using the resulting
This paper addresses a critical challenge in deploying new decision policies: the cold-start problem for prediction models. By leveraging counterfactual simulations and Sim2Real transfer, ML engineers can develop and evaluate models for new policies even before real-world data is available, significantly accelerating deployment cycles and reducing risks.