APMCM (Asia-Pacific Mathematical Contest in Modeling) 2024

Project Overview

Pet-industry modeling across China and global markets: - Built connected models to analyze domestic trends, forecast global pet-food demand, and assess tariff scenarios. - Chosen methods matched data behavior: linear trend for cats, non-linear dynamics for dogs, ARIMA for global demand, and a scenario model for export tariffs and domestic absorption.

What I Did

• Data preparation and assumptions: cleaned gaps/outliers; normalized drivers (income, urbanization, penetration) and set stable-macro assumptions.
• China pet trends: linear regression for the cat population; random forest for the dog population to capture non-linearities and feature importance.
• Global demand: ARIMA forecasting on pet-food time series to produce 3-year demand outlook and confidence bands.
• Exports & policy: scenario model for tariff shocks (e.g., 5%/10%/20%) and a domestic-market potential calculator to test offset capacity.
• Evaluation: sensitivity checks on key drivers; error diagnostics for ARIMA and RF; stress-test with alternative adoption ramps.
• Delivery: concise exhibits and an executive storyline connecting model outputs to go-to-market and policy recommendations.

Reflection

- Match method to signal: linear for steady trends; tree models for non-linear drivers; ARIMA for stationary time-series. - Assumptions are part of the model: make them explicit and stress-test with scenario ranges, not single points. - Decision focus beats metric chasing: tie every chart to a “so-what”—pricing power, export risk, domestic substitution, and timing. - For policy questions, combine models: forecasts + scenario analysis + unit-economics give a tractable decision surface for stakeholders.