M5 Department Sales Forecasting

Forecast retail department-level sales based on the M5 forecasting competition data. 7 departments tracked over 1,941 time periods.

Dataset Exploration

The M5 Department dataset is based on the M5 forecasting competition and contains Walmart department-level sales data. Features include CPI, holiday events (sporting, cultural, national, religious), SNAP benefits by state, day of week, and month.

Panel data structure: 7 department aggregations tracked over 1,941 days (13,587 total rows, 13 columns). Each department (e.g. FOODS_1) is tracked daily.

Fit Data (first 5 rows)

agg_id	ds	CPI	snap_CA	snap_TX	snap_WI	wday	month	Sales
FOODS_1	40572	1	0	0	0	1	1	2343
FOODS_1	40573	1.04	0	0	0	2	1	2216
FOODS_1	40574	1.053	0	0	0	3	1	1657
FOODS_1	40575	1.066	1	1	0	4	2	1508
FOODS_1	40576	1.075	1	0	1	5	2	1209

Forecast Data (first 3 rows)

agg_id	ds	CPI	wday	month	Sales
FOODS_1	42513	1.201	3	5	3246
FOODS_1	42514	1.201	4	5	3270
FOODS_1	42515	1.201	5	5	3274

Download sample datasets from GitHub

Code Walkthrough

Step 1: Initialize

Python

import pandas as pd
from datfid import DATFIDClient

client = DATFIDClient(token="your_DATFID_token")

Step 2: Fit the Model

Python

url_fit = "https://raw.githubusercontent.com/datfid-valeriidashuk/sample-datasets/main/M5_Department.xlsx"
df = pd.read_excel(url_fit)

result = client.fit_model(
    df=df,
    id_col="agg_id",
    time_col="ds",
    y="Sales",
    current_features="all",
    filter_by_significance=True
)

Step 3: Forecast

Python

url_forecast = "https://raw.githubusercontent.com/datfid-valeriidashuk/sample-datasets/main/M5_Department_forecast.xlsx"
df_forecast = pd.read_excel(url_forecast)

forecast = client.forecast_model(df_forecast=df_forecast)

Analysis Results (Model Fit)

Formula

Sales ~ α1*Intercept + β1*CPI + β2*Sporting + β3*Cultural + β4*National + β5*Religious + β6*snap_CA + β7*snap_TX + β8*snap_WI + β9*wday + β10*month

Alpha Estimates (Time-Invariant)

Variable	Estimate	T-stat	Interpretation
Intercept	-24,014.8	518.5	Baseline department-level sales when every driver sits at zero.

Beta Estimates (Time-Varying)

Variable	Estimate	T-stat	Interpretation
CPI	+25,321.8	45.2	A 1-unit CPI increase lifts sales by ~25,321.8 units — inflation-linked retail pricing flowing through to revenue.
Sporting	+51.1	0.4	Sporting-event days are not statistically distinguishable from non-event days in this run.
Cultural	-205.3	2.3	Cultural holidays subtract ~205.3 units from sales (stores reduce hours or shoppers stay home).
National	-758.6	9.7	National holidays subtract ~758.6 units from sales (closures and reduced traffic).
Religious	-97.6	1.3	Religious holidays move sales slightly down, but the effect is not statistically significant.
snap_CA	+212.0	7.1	SNAP days in California add ~212.0 units of sales over a typical day.
snap_TX	+256.7	7.8	SNAP days in Texas add ~256.7 units of sales over a typical day.
snap_WI	+242.7	7.4	SNAP days in Wisconsin add ~242.7 units of sales over a typical day.
wday	-242.3	38.6	Each step from Monday toward Sunday subtracts ~242.3 units — weekends still sell more, but the day-of-week index treats Saturday as the high-sales anchor.
month	-16.4	4.5	Each later month of the year shaves ~16.4 units off sales — a mild seasonal drift baked into the average.

Model Performance

0.0214

R² Overall

≈0

R² Between

0.230

R² Within

3,821.1

MAE

2.91e7

MSE

The M5 dataset has high variance across departments (FOODS vs HOBBIES vs HOUSEHOLD), which explains the lower between-group R². Adding lagged features or filtering by significance can improve results.

Try it yourself: Select "M5 Department" in the Free Playground.