Much business data is recorded over time. Analysing it well means separating signal from calendar noise.
Plot First
A simple line chart reveals trends, seasonality, outliers, gaps and changes in level that statistics alone can hide.
Decomposition
Split a series into trend, seasonal and residual components. Classical decomposition and STL (seasonal-trend decomposition using LOESS) are standard tools.
Calendar Effects
- Months have different numbers of days and weekends.
- Public holidays move between years (Easter, lunar new year).
- Trading-day differences can create apparent growth.
Adjust for these before comparing periods.
Comparing Periods Fairly
- Year over year comparisons remove regular seasonality.
- Rolling averages smooth noise (a 7-day average for daily data with weekly patterns).
- Seasonally adjusted series allow month-to-month comparisons.
Growth Rates
Percentage changes from a small base look dramatic; show absolute values alongside. Compound annual growth rate summarises growth over several years.
Structural Breaks
Changes in definitions, systems or business model create jumps that aren't real changes in behaviour. Annotate them on charts and handle them in analysis.
Autocorrelation
Consecutive values in a time series are related. Standard statistical tests that assume independent observations can give misleading results on time series.
Tools
pandas for resampling and rolling windows; statsmodels for decomposition and time-series models.