ggbond0905/24-7-market-volatility-analysis

How Volatile Is a 24/7 Market? A Simple Data Experiment

Traditional financial markets have a familiar rhythm. They open, trade for several hours, and eventually close. Digital asset markets work differently: activity continues around the clock, across countries, time zones, and weekends.

This dataset project explores a simple question: does a market that never closes behave the same way throughout all 24 hours?

At first, “24/7 trading” can make every hour sound interchangeable. In practice, availability and activity are two different things. Participation changes throughout the day, liquidity can shift, and market activity may rise or fall as different regions become active.

The purpose of this project is not to predict prices or identify the “best” trading hour. Instead, it demonstrates a simple way to organize hourly observations and examine volatility using Python.

What Do We Mean by Volatility?

Volatility sounds more complicated than it needs to be. At a basic level, it describes how much prices move over a given period.

Imagine that a market rises 0.8% during one hour and falls 0.8% during another. The direction is different, but the size of the movement is identical.

For this experiment, I use absolute percentage movement as a simple measure. A return of +0.8% and a return of -0.8% are therefore both treated as a movement of 0.8%.

This lets us ignore direction for a moment and focus on a different question: how large are the movements occurring during each hour of the day?

Data Structure

Market information is commonly organized into candlestick data containing open, high, low and close prices for a particular interval.

While looking into how this information is structured programmatically, I reviewed the BYDFi API documentation as a reference for market endpoints and candlestick data.

For this project, however, I decided to work with simulated hourly observations rather than data pulled directly from a live API. This keeps the experiment simple and reproducible without requiring API credentials or depending on a live service.

The trade-off is important: simulated data cannot tell us how a real asset actually behaves. The dataset demonstrates an analytical workflow rather than documenting a real trading pattern.

Dataset Design

The example dataset represents 30 days of hourly observations, giving us 720 timestamps in total.

Each observation can be organized around three simple fields:

• timestamp: the date and hour of the observation
• return_pct: the simulated percentage movement during that hour
• abs_move: the absolute size of that percentage movement

We can also extract the hour from each timestamp, producing values from 0 to 23.

Once the observations are grouped by hour, we can calculate the average absolute movement for each group.

Conceptually, the workflow looks like this:

Hourly observations → Percentage movement → Absolute movement → Group by hour → Compare averages

This reduces hundreds of individual observations to 24 hourly categories that are much easier to compare.

A Simple Python Workflow

A minimal implementation only requires pandas, NumPy and Matplotlib.

The process is straightforward:

1. Generate or load timestamped hourly observations.
2. Calculate the percentage movement for each observation.
3. Convert each movement to its absolute value.
4. Extract the hour from each timestamp.
5. Group the observations from hour 0 through hour 23.
6. Calculate the average absolute movement for each hour.
7. Visualize the result as a bar chart.

The result is a simple 24-bar visualization showing how the average magnitude of movement differs across hourly groups.

What Can We Learn From It?

With simulated observations, some hourly averages will naturally be higher than others.

However, there is an important limitation: those differences should not be interpreted as evidence that a particular hour is genuinely more volatile in real markets.

Randomly generated observations contain random variation. A tall bar in this experiment demonstrates how the analysis works, not a persistent market phenomenon.

What is more useful is the workflow itself.

We start with hundreds of timestamped observations, transform them into a consistent measure, group them according to time, and reduce the result to something that can be inspected visually.

The same approach could later be applied to real historical OHLC data.

Extending the Experiment

With a larger real-world dataset, the analysis could explore several additional questions:

• Are certain hours consistently associated with larger price movements?
• Do patterns change when major financial centers become active?
• Are weekends noticeably different from weekdays?
• Is higher trading volume associated with larger movements?
• Do the same hourly patterns appear across different assets?

Instead of looking at only 30 days, a more serious analysis could cover six months, one year, or several years.

Longer periods are particularly useful because a pattern appearing during a few weeks may disappear once more observations are included.

Volume could also be added as another variable. This would make it possible to compare market activity and price movement instead of treating volatility in isolation.

Limitations

This project is intentionally simple.

The observations are simulated, so the dataset should not be used to draw conclusions about the historical behavior of any particular asset.

Absolute percentage movement is also only one way to describe volatility. More detailed studies might use standard deviation, rolling volatility, high-low ranges, realized volatility, or other statistical measures.

Time zones require care as well. Before comparing hourly activity from a real dataset, timestamps should be standardized, typically to UTC, so observations are grouped consistently.

These limitations are part of the point of the experiment: a chart is only as meaningful as the assumptions and data behind it.

24/7 Does Not Mean 24 Identical Hours

The central idea is simple: a market being available 24 hours a day does not mean human activity is evenly distributed across those 24 hours.

Liquidity can change. Participation can change. Trading volume can change. News and economic events occur at particular times, while different regions of the world become active and inactive throughout the day.

So “always open” and “always equally active” are not the same thing.

This small dataset provides a starting point for examining that idea with data rather than assumptions. A future version could replace the simulated observations with real hourly candlestick records, compare multiple assets, incorporate volume, and separate weekdays from weekends.

Sometimes a useful data project can begin with something surprisingly small: ask one clear question, find a way to measure it, and see what the data reveals.

This dataset is intended for educational and data-analysis purposes only. It is not financial or investment advice.