Walmart Sales (Repo)

Project Description: Utilizing data provided by Walmart via Kaggle available here, I aggregate sales by store level and perform exploratory analysis of weekly store sales and SARIMA models to predict future sales as time series.

EDA of Weekly Sales

Before generating any formal models, I explore the completeness of data coverage and explore trends in sales over time, detailed in my notebook hosted here.

Insight 1: Walmart store sales display an annual cycle with a massive spike during Christmas season and decline thereafter.

Insight 2: Superbowl Sunday and Labor Day do not stand out visually while the Black Friday spike and post-Christmas crash are massive.

Insight 3: Viewing the average autocorrelation values across all stores, only the first lag appears statistically significant with some evidence of a monthly cycle. Additionally, the 52-week lag correlates more strongly with future weekly sales than any other time lag.

Insight 4: Out of 45 stores, 8 fail the ADFuller test of stationarity. I find 5 stores display consistent, long-term trends while 3 display evidence of temporary shocks within the provided timeframe.