Riding the AI Wave: Forecasting NVIDIA (with AMD & Intel)

Project Overview

This project models monthly closing prices (2020–2025) for NVIDIA and compares them with AMD and Intel. I performed stationarity checks (ADF), variance stabilization (Box–Cox), differencing, ACF/PACF diagnostics, and fitted ARIMA family models. The selected models were then used to generate 12-month forecasts and quantify uncertainty bands. Insights connect statistical signals with real market narratives in the AI chip cycle.

What I Did

• Owned end-to-end NVIDIA pipeline: data sourcing/cleaning, EDA, and ADF stationarity tests.
• Applied Box–Cox (λ≈−0.39) and first-order differencing; verified stationarity via diagnostics.
• Compared ARIMA (1,1,0), (0,1,1), (1,1,1), (2,1,0), (0,1,2); selected ARIMA(0,1,0) on AIC and parsimony.
• Checked residuals (ACF/PACF within bounds), ran Ljung–Box; QQ plot acceptable for financial data.
• Forecasted 12-month horizon on transformed scale and inverse-transformed back to dollars with 95% CI.
• Benchmarked NVIDIA against AMD/Intel to contextualize momentum vs. volatility.
• Translated signals into stakeholder-friendly narrative on AI-cycle demand and competitive pressure.

Reflection

Working exclusively on the NVIDIA track taught me to own the full lifecycle of a data science project: from raw data to interpretation. I realized how small statistical choices (differencing order, Box–Cox parameter) dramatically change NVIDIA forecasts. Producing NVIDIA’s forecast charts reinforced that communicating uncertainty bands is as crucial as the forecast itself. Most importantly, I learned to translate NVIDIA’s time-series signals into a market story that resonates with non-technical stakeholders.