Untangling the complexity of demand forecasting models: building a Market Simulator
summary
A working Market Simulator is built in Python (numpy, pandas and an ensemble approach) that models expected demand as a price‑dependent curve, samples Poisson sales realizations, and trains regressors to forecast profit; exploration uses upper confidence bound style optimism realized via bootstrap quantiles (30 models) to bound uncertainty. A deployed pricing model claims a jump from an expected 1% margin lift target to a delivered 28% improvement. Regret stays under 5% across scenarios and the simulator targets fast iteration (three‑month improvement cycle).
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:48Pricing optimization motivation
- 3:59Simulate market for forecasting
- 5:24Simulator assumptions and scope
- 7:08Demand curve modeling
- 8:10Profit optimization target
- 11:00Market simulation loop
- 12:22Exploration in slow markets
- 14:30Multi-arm bandits for exploration
- 17:42Bootstrapping confidence intervals
- 19:48Model validation
- 21:54Simulation results and regret
- 23:19Closing remarks and limits
- 24:24Q&A competitor markets
- 28:38Q&A pricing factors
- 29:40Q&A sparse data
- 31:25Q&A technical implementation
related talks
28mInterpretable and realistic generative models in data science?
28mDemand Forecasting: Accuracy, Stability, Automation
59mPredictability Check: Demand Forecasting
29mProbabilistic demand forecasting at scale
30mClimbing the causal ladder for fun and profit
01h 18msupplyseer: Computational Supply Chain with Python
25mParadoxes in model training and evaluation under constraints
31mBuilding an electricity market model from scratch