Stop 09 · Advanced Analytics & Predictive BasicsTopics · 4/8

Time Series Analysis Basics

In short

Sales data isn't just numbers, it's numbers in order. Time adds trend, seasons and sudden shocks, and comparing this October with last October usually tells you more than comparing it with September.

Analyzing data points ordered over time — trend, seasonality, and noise — the foundation for any forecasting or 'is this metric actually growing' question.

The jaadu pours chai into a long row of glasses from sunrise to evening.

Think of it like this

A chai stall through the day

The chai wala knows the rush: busy at 8 AM, quiet after lunch, busy again at 5. Over the years, the crowd grows too. That daily rhythm is seasonality, the slow growth is trend, and a random rainy day is noise. Time series analysis separates the three.

Watch it happen

Monthly orders, April to November

100

104

103

108

112

118

155

165

1/5Eight months of orders in thousands. Read left to right, it's a story in time order.

The big ideas

What you need to know

Trend

The long-run direction, up, down or flat. Rolling averages smooth noise so you can see it.

Seasonality

Patterns that repeat on a calendar: weekends, month-end salary days, festivals, exam season.

Year-over-year

Comparing the same period across years removes seasonality from the comparison.

Shocks and breaks

Lockdowns, price changes or an app outage. Mark them, or your trend will lie.

See it in code

Walk through the code

python

1s = (2    df.set_index("order_date")["orders"]3      .resample("MS").sum()4)5 6trend = s.rolling(3).mean()7yoy = s.pct_change(12)8 9from statsmodels.tsa.seasonal import seasonal_decompose10parts = seasonal_decompose(s, model="additive", period=12)11parts.plot()

1/4Put dates in the index and add up orders per calendar month. MS means month start.

Monthly totals, a 3-month rolling average, year-over-year growth and a trend plus season breakdown. Decomposition needs at least two full years.

Save yourself the pain

Common mistakes

Oops

Reporting 'November is up 6% on October' as great news.

Do this instead

Festive months are always higher. Compare with last November to see actual growth.

Oops

Leaving missing months out, so the series silently skips time.

Do this instead

Resample to a regular frequency so gaps show up as zeros or NaN that you can handle.

Your mission

Find the seasons

~30 min

Quick check

A gift shop's December sales are always the highest. What's the fairest way to judge this December?

Go deeper

1Plot a time series and visually identify trend versus seasonal patterns.
2Learn the difference between a trend, a seasonal pattern, and pure noise.
3Try a simple moving average to smooth out noise in a noisy time series.