Time series + forecasting for accounting data

Why this exists: Forecasting becomes less scary once you’ve built clean daily/monthly series. This chapter outlines the progression.

Learning objectives

• Explain trend, seasonality, and noise using accounting time series.

• Build a baseline forecast and evaluate it.

• Understand when forecasting is inappropriate (garbage in / structural breaks).

Outline

Start with baselines

• Start from normalized/gl_journal.csv and build a clean daily/monthly series (revenue proxy, expense totals, or cash).

• Last value, moving average, seasonal naive.

• Always do a simple backtest (train on earlier months, test on later months).

• Compare forecasts with simple error metrics.

Add explanatory variables

• Promotions, holidays, payroll cycles, or other known drivers.

• Use regression as a driver model (not magic).

Keep it business-grounded

• Always interpret: what would make the forecast wrong?

• Document assumptions and data limitations.

• Structural breaks examples: pricing changes, a new location, system migrations, one-time events, policy changes.

Where this connects in the workbook

• Track D chapter index (PyPI) (chapters that introduce forecasting ideas)

• Track D: Apply what you learned to your data (how to apply the same methods to your exports)

Note

This page is intentionally an outline right now. Expand it incrementally as we refine Track D narrative.