Spreadsheets are flat. The economy is not.
I’ve been working on a prototype to make macroeconomic data less painful to look at and a lot more fun to explore. Instead of staring at endless rows and columns, you can fly through it as a 3D network.
Live demo here: monthly-economic-networks-pe.vercel.app
Screenshots:
The data
Built from the Central Bank of Peru’s (BCRP) monthly time series. In the [main] Variable-Metadata bigraph, we show a total of more than 7,000 nodes and more than 90,000 edges, from which 7105 of the nodes are macroeconomic variables (time series) and the remaining nodes are metadata groups.
The graph
A bipartite network:
- variables on one side
- metadata groups on the other
Edges = variable membership in a metadata group.
Features
- Switch between 2D radial and 3D layouts
- Inspect nodes for labels, connections, and metadata
- Light/dark mode, zoom, pan, fly-through
- Node scaling by connectivity
- Missing metadata shown (because real-world data is never perfect, and this makes it visible)
Why bother?
On one hand, this is just a visualization playground. On the other, it’s part of a proposal: applying AI to automatically select and explain macroeconomic variables through metadata connections. The bigger idea is to give researchers and institutions tools to see relationships in data, not just crunch them.
4 Likes
Yep, a carefully crafted visualization can turn a mess of numbers into a meaningful story.
Unfortunately, I have no economics education (neither macro, nor meso or micro), so I cannot estimate the usefulness of a macroeconomics 3D network. Besides variables-vs-metagroups, is there any other spacial meaning of nodes locations? Here are some ideas, very raw ideas, feel free to ignore them:
- it might be good to arrange nodes so that distance means something – semantic closeness or economics correlation
- color could be used to indicate closeness, for example, clicking on a node colors its immediate neighbours in one color, their neighbours in another color and so on
- visualizing a path between two nodes might make some meaning as it connects two variables via other variables
- when node neighbours are shown it might be good to show short labels above their balls
- some predefined filtering may reduce the visual complexity of the network (e.g. show all node with some propery, and hide the others or make them transparent)
2 Likes
Thanks a lot for the suggestions, Pavel. At the moment this is a bipartite graph, so edges only link variables to their metadata groups. The layout doesn’t carry learned meaning (like correlations or distances), so ideas such as semantic spacing or multi-hop coloring don’t quite fit yet.
That said, filtering, neighbor highlighting, and labels are really useful, and I’ll keep your other ideas in mind for a future correlation-based graph where they’d shine. Appreciate your input!
2 Likes
I think it’s only missing an autoplay feature. That could highlight the critical tables in descending order. Which, after imposing a compact spiral, relieves the user’s overwhelming need to click an obscure orb. So the user could leisurely enjoy the information in this engaging and interpretive format… But potentially pivot to a more detailed dashboard. Otherwise analytics for engagement would be useless since clicks are naive and nondeterministic.
1 Like
