DATFID Documentation

The Real Power of DATFID: Panel Data Analysis

DATFID is purpose-built for panel data analytics — longitudinal data where multiple entities (products, stores, customers, regions) are observed over time. This is fundamentally more powerful than analyzing individual time series because it combines the cross-sectional dimension (differences between entities) with the time dimension (changes over time).

What Is Panel Data?

Panel data is a dataset where the same entities (e.g. products, stores, customers) are observed at multiple time points. A typical panel dataset has:

• An entity ID column — identifies each product, store, or customer
• A time column — the observation timestamp
• A target variable — the value you want to predict (revenue, demand, risk, etc.)
• Feature columns (optional) — additional variables like price, promotions, demographics

For example: tracking Revenue for 50 products over 24 months, or Loan Probability for 1,000 customers over 12 quarters.

Benefits for Real Business Use Cases

• More information, better estimates: By pooling data across entities and time, panel models produce more reliable coefficients than analyzing each entity separately.
• Time-invariant parameters (alphas): This is a key advantage normally missed by standard analytics. DATFID estimates alpha coefficients — parameters that capture the inherent, time-invariant characteristics of each entity. For example, a store's baseline revenue level or a customer's inherent risk profile. These alphas tell you "how does this entity differ from others on average, independent of time?" Most tools ignore this entirely; DATFID surfaces it.
• Time-specific effects: Capture macro trends that affect all entities simultaneously (seasonality, economic shifts).
• Cross-sectional comparisons: Understand which products, stores, or customers behave differently and why.
• Scalability: Natural fit for revenue, demand, risk, and performance forecasting across hundreds or thousands of entities.
• Interpretability: Every coefficient has a clear business meaning — you get a formula, not a black box.

Supported File Formats

DATFID accepts Excel (.xlsx, .xls) and CSV (.csv) files for both model fitting and forecasting. Sample datasets are available in the Free Playground and on the sample-datasets GitHub repo.

Why Do You Need a Separate Forecast File?

DATFID's workflow has two steps:

1. Fit step: You send historical data (entity + time + target + features) to train the model. The result is a formula with interpretable coefficients.
2. Forecast step: You send a forecast file with the same structure, but covering the periods you want to predict. This file defines which entities and time points to generate predictions for — it is not additional training data, but a "request" for predictions.

Each sample fit dataset has a matching forecast file (e.g. Food_Beverages.xlsx → Food_Beverages_forecast.xlsx).

Get Started

Use Cases

Explore real-world examples with sample datasets, analysis results, and forecast outputs.