---
title: "What Is Time Series Data? Examples and Analysis"
description: "Time series data is a set of values recorded at successive points in time, such as daily prices or server metrics. Examples, components, storage and analysis."
url: https://proxynet.io/blog/what-is-time-series-data
date: 2026-11-19
author: "Acar Diveroli"
category: "Use Cases"
lang: en
---

# What Is Time Series Data? Examples and Analysis

A pricing analyst exports a year of one competitor's daily prices for a coffee grinder: 365 rows, each with a date and a number. Sorted by price, the file says almost nothing. Sorted by date, it shows a slow climb through the spring, a drop in the last week of November, and one morning in early November when the price reads 0.99. Is the product getting more expensive, is November always cheaper, and was that 0.99 a real offer or a broken page?

Those three questions are what time series data is for. This post covers what makes data a time series, how it differs from cross-sectional and panel data, its four components, stationarity in plain words, collection and storage, the basics of analysis and forecasting, and the mistakes that quietly ruin a series.

> **Note: Short answer**
>
> Time series data is a sequence of values of the same thing measured at successive points in time, stored with a timestamp and kept in time order: a product's price every day, a server's CPU load every 15 seconds, a country's inflation rate every month. The order carries the information: analysis looks for trend (the long-term direction), seasonality (a pattern that repeats on the calendar), cycles and noise, and uses them to monitor, compare and forecast.

## What is time series data?

Time series data is a set of observations of one variable, each tied to the moment it was measured and arranged in time order. Every row has at least two parts, a timestamp and a value, and usually a third: an identifier that says what was measured (a product, a sensor, a server).

The NIST/SEMATECH e-Handbook of Statistical Methods defines a time series as [an ordered sequence of values of a variable at equally spaced time intervals](https://www.itl.nist.gov/div898/handbook/pmc/section4/pmc41.htm). "Ordered" is the key word: shuffle the rows and the meaning is gone, because yesterday's price explains today's in a way no random row does. "Equally spaced" is the classic textbook case; in practice many series arrive at uneven moments, as a later section explains.

A short extract from the grinder example shows the shape:

| Time (UTC) | Product | Price |
|---|---|---|
| 2026-11-04 09:00 | Grinder X | 26.50 |
| 2026-11-05 09:00 | Grinder X | 24.90 |
| 2026-11-06 09:00 | Grinder X | 0.99 |
| 2026-11-07 09:00 | Grinder X | 24.90 |

One product, one value per day, one fixed time of day. Add a second product and you have two series in one table; the product column keeps them apart.

## What are examples of time series data?

Anything measured repeatedly over time forms a time series. The most common examples:

- **Prices:** a product's shelf price per day, a hotel rate per night, a currency exchange rate at the daily fixing.
- **Server metrics:** CPU load, memory use, requests per second and error counts sampled every few seconds.
- **Sensor readings:** temperature, energy use or machine vibration logged by devices.
- **Business activity:** daily visits, orders per hour, stock levels per warehouse.
- **Economic statistics:** monthly inflation, quarterly GDP, the monthly unemployment rate.
- **Weather:** hourly rainfall, daily air quality, yearly average temperature.

What they share is the question behind them: how does this value change, and what will it be next?

## What is the difference between regular and irregular time series?

A regular time series has a value at fixed intervals, such as every minute or every day; an irregular one has values only when something happens. A temperature sensor that reports every 60 seconds produces a regular series. Trades on an exchange, error events in a log, or a price recorded only when it changes produce irregular ones, with three values one minute and none for an hour.

The interval is called the frequency of the series: per second, hourly, daily, monthly. Many tools expect a regular series, so irregular data is resampled into fixed buckets, and how you fill each bucket is a real decision: carry the last known price forward, sum the sales, average the temperature. An empty bucket needs a rule too, because "no reading" and "zero" are different facts.

## How is time series data different from cross-sectional and panel data?

The difference is what varies: time series data follows one subject across many moments, cross-sectional data compares many subjects at one moment, and panel data follows many subjects across many moments. The same price file can be cut all three ways:

| Data type | What varies | Example | Typical question |
|---|---|---|---|
| Time series | Time | One grinder's price, daily for a year | Is it rising? |
| Cross-sectional | Subjects | 200 grinders' prices on one day | Who is cheapest today? |
| Panel (longitudinal) | Subjects and time | 200 grinders, daily for a year | Who cuts first, who follows? |

Panel data is the most informative and the most demanding: every subject needs a stable identifier, otherwise the same product appears as two subjects after a page redesign.

## What are the components of a time series?

Most time series can be described as four components added or multiplied together: trend, seasonality, cycles and noise. Separating them is the first step of almost every analysis, because each answers a different question.

- **Trend** is the long-term increase or decrease. It need not be a straight line; a price can climb steeply and then flatten.
- **Seasonality** repeats with a fixed, known period tied to the calendar: weekend dips in office traffic, a December peak in retail sales.
- **Cycles** are rises and falls without a fixed period, such as business cycles that last several years and vary in length.
- **Noise** (the remainder) is what is left: random variation, one-off events and measurement errors.

Seasonal and cyclic are the pair people mix up most: if the pattern returns on a calendar schedule, it is seasonal; if its timing cannot be predicted, it is a cycle. The open textbook *Forecasting: Principles and Practice* by Rob J Hyndman and George Athanasopoulos explains the [trend, seasonal and cyclic patterns](https://otexts.com/fpp3/tspatterns.html) with plotted examples and is free to read online.

## What does stationarity mean in a time series?

A stationary time series is one whose statistical behaviour, such as its average level and how much it swings, does not depend on when you look at it. Pick any stretch of a stationary series and it looks broadly like any other stretch.

A series with a trend is not stationary, because its level depends on the date; nor is a seasonal one, because December behaves differently from March. A series with cycles but no trend or seasonality can still count as stationary, because the timing of its peaks cannot be predicted.

It matters because many classic forecasting models assume stationarity, so analysts transform a series first. The usual tool is differencing: each value is replaced by its change since the previous observation, and "up 0.40" is often more stable than "26.90". A logarithm helps when the swings grow with the level, as they do for many prices.

## How is time series data collected?

Time series data is collected by measuring the same thing, the same way, at planned times, and writing each value down with its timestamp. Whether the source is a sensor, a server or a public web page, a reliable job follows five steps:

1. **Define the measure.** Decide exactly what one value means: the price shown to a guest or a logged-in buyer, with or without shipping, in which currency.
1. **Pick the frequency.** Match it to how fast the value changes. Daily is plenty for most shelf prices; server metrics need seconds.
1. **Timestamp in UTC.** Store each value with an unambiguous time and its offset, in the format [RFC 3339](https://www.rfc-editor.org/rfc/rfc3339) defines for internet timestamps: `2026-11-06T09:00:00Z`, where `Z` means UTC. Convert to local time only for display.
1. **Keep the identifier stable.** Use a product code or URL that survives redesigns, so the series does not split in two.
1. **Record how the value was obtained.** Note the method, location and any change to it, so a jump caused by your own setup is not mistaken for a real one.

For web sources, a monitor that checks pages on a schedule already produces a time series of page versions; [How to Monitor a Website for Changes](/blog/website-change-monitoring) shows how such checks are set up and how to keep the load on the site low.

## How is time series data stored?

Small time series fit comfortably in a spreadsheet or an ordinary relational database with an index on the timestamp. Large ones, with thousands of series and new values every few seconds, usually go into a time-series database, a database designed for data that arrives in time order and is mostly appended, rarely edited.

Time-series databases store data in chunks by time range, so a query for "last 24 hours" reads only the newest chunks; they compress neighbouring values, which are often similar; and they make old data cheap to drop. The open-source monitoring system Prometheus is a well-documented example: its [storage documentation](https://prometheus.io/docs/prometheus/latest/storage/) says it groups incoming samples into two-hour blocks, compacts them into longer blocks in the background, and keeps data for 15 days by default.

Two terms come up in every storage plan. **Retention** is how long raw data is kept before deletion. **Downsampling** replaces old high-resolution data with summaries, for example one-minute values for two weeks and hourly figures for a year. Keep the minimum and maximum next to the hourly average, because an average hides a five-minute outage.

## How do you analyze time series data?

You analyze a time series by plotting it first, smoothing it to see the trend, separating out seasonality, and only then measuring or forecasting. Skipping the plot is a costly shortcut, because a chart shows gaps, spikes and level shifts that summary numbers hide.

**Moving average.** A moving average replaces each point with the average of the last few points, which smooths out day-to-day noise. It also shows why cleaning comes first. In the grinder data, where the price was 24.90 on 2 and 3 November, the 0.99 reading drags the three-day average down for three days:

| Date | Price | 3-day average |
|---|---|---|
| Nov 4 | 26.50 | 25.43 |
| Nov 5 | 24.90 | 25.43 |
| Nov 6 | 0.99 | 17.46 |
| Nov 7 | 24.90 | 16.93 |
| Nov 8 | 23.50 | 16.46 |

A moving median would barely notice the outlier; better still, check the page and decide whether 0.99 was real before it enters any average. Cleaning steps for collected data, from text to types and outliers, are covered in [How to Clean Scraped Data with Pandas](/blog/clean-scraped-data-with-pandas).

**Decomposition.** Decomposition splits a series into trend, seasonal and remainder parts so you can look at each one. Statistics offices publish many indicators "seasonally adjusted": Eurostat describes [seasonal adjustment](https://ec.europa.eu/eurostat/statistics-explained/index.php?title=Glossary:Seasonal_adjustment) as removing the effects of recurring seasonal influences so that the underlying movement is easier to see. That is why a December jump in retail sales may vanish from the adjusted figure.

**Forecasting.** Forecasting estimates future values from past ones. Start with two baselines: the naive forecast (tomorrow equals today) and the seasonal naive forecast (next Monday equals last Monday). Exponential smoothing gives recent values more weight, ARIMA models predict from past values and past errors of the differenced series, and machine learning models can add outside drivers such as promotions. A complex model earns its place only if it beats the baselines on a recent period you held back.

## Why do time zones break time series?

Time zones break time series because local clocks skip and repeat hours, so a "day" in local time is not always 24 hours long. On our test machine, Node.js 24 with its built-in time-zone data reported that 25 October 2026 in Berlin lasts 25 hours and 29 March 2026 only 23, and two moments an hour apart, 00:30 and 01:30 UTC on 25 October, both display as 02:30 local time.

Store local times without an offset and that autumn night produces two readings with the same timestamp, while the spring night has an hour with no readings at all; daily totals come out an hour too long or too short. The fix is the collection rule: store UTC with an explicit offset and convert to local time only when a report groups by local day.

## Where do proxies fit in collecting time series data?

Proxies matter when the time series you need comes from public web pages that differ by country. A retailer often shows prices, stock and delivery options for the visitor's location, so a price series collected from your office records your office country's version every day, not the market you are studying.

The fix follows the collection rules above: request the same pages, at the same interval, from the same country every time, so that only the price changes in the series. A [Residential Proxy](https://proxynet.io/residential-proxy) can send each check through a home connection in the country, or the city, the series describes, at a rate the site can easily handle.

Our [price monitoring](/price-monitoring) solution covers tracking prices, stock status and minimum advertised price compliance at regular intervals.

A proxy does not change what you may collect: the site's terms and robots.txt still apply. The practical side of matching products and choosing check intervals is covered in [How to Track Competitor Prices](/blog/competitor-price-tracking).

## Where is time series data used?

Time series data is used wherever a decision depends on how something changes over time:

- **E-commerce:** price history, stock levels, demand forecasts for purchasing.
- **Finance:** share prices, exchange rates, risk measured over rolling windows.
- **IT operations:** dashboards and alerts on server metrics.
- **Energy:** load forecasts, meter readings, solar and wind output.
- **Manufacturing:** machine sensors that reveal wear before a breakdown.
- **Economics:** inflation, employment and trade statistics.

## Common mistakes with time series data

Most time series problems are created at collection time and only discovered at analysis time:

- **Local timestamps without an offset.** Duplicate and missing hours at every daylight saving change.
- **Treating a gap as zero.** A page that failed to load did not show a price of zero.
- **Changing the method mid-series.** A new source, location or definition creates a step that looks like a real change. Mark the date.
- **Comparing a seasonal month with the previous one.** Compare December with last December, or use adjusted figures.
- **Smoothing before cleaning.** One bad value distorts every average it touches.
- **Testing a forecast on data it has already seen.** Hold back the most recent period.
- **Downsampling away the evidence.** Averages hide spikes; keep minimum and maximum.

## Decision guide

| Need | Recommendation |
|---|---|
| A few series, updated daily | A spreadsheet or relational table with a time index |
| Thousands of series, values every few seconds | A time-series database with retention rules |
| See the direction under noisy data | Moving average or median |
| Compare months fairly | Seasonally adjusted figures or same month last year |
| Forecast next month | Seasonal naive baseline first, then smoothing or ARIMA |
| Data from several time zones | Store UTC with offset, convert for reports |
| Web prices that vary by country | Collect from that country at a fixed interval |
| Irregular events into a chart | Resample into fixed buckets with an explicit fill rule |

## Frequently asked questions

### What is time series data in simple terms?

Time series data is a list of measurements of the same thing taken one after another over time, each with its date and time. A daily price history, a monthly electricity bill and a fitness tracker's heart rate per minute are all time series.

### What is the difference between time series and panel data?

Time series data follows one subject over time, while panel data follows many subjects over the same period. One product's daily price for a year is a time series; 200 products' daily prices for the same year form a panel, which lets you compare how different subjects change.

### What does seasonally adjusted mean?

Seasonally adjusted means the regular calendar pattern has been removed from a series so that the underlying movement is easier to see. A seasonally adjusted retail figure does not jump every December, so a rise in it signals a real change rather than the holiday season.

### Do I need a time-series database?

No, not for a small number of series. A spreadsheet or a regular database table with a timestamp index handles years of daily data easily. A time-series database starts to pay off when you write many values per second across many series and need retention and downsampling rules.

### What is a stationary time series?

A stationary time series has statistical properties, such as its mean and variance, that stay the same over time. Series with a trend or seasonality are not stationary; analysts often make them stationary by differencing, that is, by working with the change from one observation to the next.

### Can I analyze time series data in Excel?

Yes, for the basics. Excel can chart a series, compute moving averages with a simple formula and, in Excel 2016 and later, build a forecast with the Forecast Sheet command on the Data tab. Its timeline needs consistent intervals, so resample irregular data first.

## Summary

Time series data is the same measurement repeated over time, stored with a timestamp and read in order. It tells you which way a number is heading, what repeats on the calendar, what is noise and what may come next. Collect it consistently, timestamp it in UTC, keep identifiers stable, clean before you smooth, and test forecasts against simple baselines. When the series comes from web pages that differ by country, collect every check from the country it describes, so that the only thing moving in your data is the value you set out to measure.
