This paper is concerned with the use of simulation in computing predictors in settings in which real-world observations are collected. A major challenge is that the state description underlying the simulation will typically include information that is not observed in the real system. This makes it challenging to initialize simulations that are aligned with the most recent observation collected in the real-world system, especially when the simulation does not visit the most recently observed value frequently. Our estimation methodology involves the use of "splitting," so that multiple simulations are launched from states that are closely aligned with the most recently collected real-world observation. We provide estimators both in the setting that the observed real-world values are discrete and are continuous, with kernel smoothing methods being systematically exploited in the continuous setting.
- Simulation-Based Prediction
- Eunji Lim - Decision SciencesPeter W Glynn - Stanford University
- Operations research, Vol.71(1), pp.47-60
- INFORMS
- 15
- Adelphi's Celebration of Scholarly Research & Creative Works; Adelphi's Celebration of Scholarly Research & Creative Works 2022; Adelphi University; Decision Sciences and Marketing; Office of the Provost; Robert B. Willumstad School of Business
- English
- Journal article
- Adelphi's Celebration of Scholarly Research & Creative Works 2022
- https://doi.org/10.1287/opre.2021.2229
- 991004281525606266