Using Simulation and Resampling to Answer Practical Statistical Questions
Summary
This article argues that programming simulations can make statistical questions more intuitive than relying solely on classical formulas. It illustrates the approach with roulette: under a stated single-number win probability, repeated simulated sequences estimate how often a player would reach a specified number of wins by chance. The result is compared with a binomial calculation, showing that simulation can approximate the analytical probability while making the experiment easier to understand.
The example also explains why simulation results vary and why more repetitions improve their empirical precision. It notes that observing no simulated extreme outcomes does not establish that the true probability is zero; a finite run may simply be too small to encounter a rare event. The author presents simulation as useful when a generative model is available, while cautioning that many trading problems offer only historical data or uncertain assumptions. The method supports intuition and inquiry, but its reliability depends on modeling choices and the number of simulations.
Key ideas
- Simulation can estimate probabilities by repeatedly generating outcomes from an explicit model.
- A roulette example compares simulated win counts with a binomial probability calculation.
- More simulation runs generally make empirical estimates converge toward theoretical probabilities.
- A zero count in a finite simulation does not mean an event is impossible.
- Trading applications may be harder because the underlying data-generating process is often unknown.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.