Time Series Basics

The one idea: Any line on a dashboard is really three things mixed together — a long-run trend, repeating calendar patterns, and random noise. Most false alarms come from mistaking a calendar pattern or an unfinished period for a real change. Always ask which of the three moved before you ask why.

⚠️ The trap: "This month's revenue is down 40% versus last month — something broke." That feels right and it's wrong. The month is only part-finished, so of course it looks like a crash; compare period-to-date against prior period-to-date (June's first 18 days versus May's first 18) before you sound any alarm.

"Signups are down 12% this week — what happened?" Half the time, the answer is: nothing happened. The week contained a holiday, or the comparison week contained a product launch, or signups are always down the first week of the month, or the week isn't over yet. Time-series questions look simple — is the line going up or down? But the line you're staring at is actually three signals braided together. Misreading which one moved is the most common way analysts raise a false alarm.

Every time series is three things at once

Any business metric over time can be thought of as:

1. Trend — the long-run direction. Is the business actually growing or shrinking, underneath everything else?
2. Seasonality — repeating, calendar-driven patterns. Weekly cycles (B2B products crater on weekends), monthly cycles (payday spikes), yearly cycles (retail Q4, travel summer). Seasonality is predictable movement — it tells you about the calendar, not the business.
3. Noise — everything else: random fluctuation, one-off events, measurement wobble. Individual days…