Ask HN: Has anyone tried to model the entire business world as a graph?
Node types being things like: companies, macro impacts (interest rates), value chain components (raw materials/trucking) Edge types being things like: suppliers, buyers, competitors
What would be the ideal node types and edge relationships to model the business world in a way you can run simulations of events?
Has this already been done? Seems like it would be helpful.
Sylvie and Bruno Lewis Carroll, 1893.
"That's another thing we've learned from your Nation," said Mein Herr, "map-making. But we've carried it much further than you. What do you consider the largest map that would be really useful?"
"About six inches to the mile."
""Only six inches!"exclaimed Mein Herr. "We very soon got to six yards to the mile. Then we tried a hundred yards to the mile. And then came the grandest idea of all! We actually made a map of the country, on the scale of a mile to the mile!"
"Have you used it much?" I enquired.
"It has never been spread out, yet," said Mein Herr: "the farmers objected: they said it would cover the whole country, and shut out the sunlight! So we now use the country itself, as its own map, and I assure you it does nearly as well.
I am not very clever so I assume this means "it won't be used"?
Sometimes StackOverflow energy bleeds into Hackernews lol
What would you use it for?
The same thing you use all maps for.
The problem with the physical map is that you can't navigate it any better than the real world. With a virtual map there's no trade-off save for the cost of making it.
See also Baudrillard's Simulacra & Simulation, referenced in "The Matrix"
This has been done in the macroeconomics literature to model bank interconnectedness and financial crises. This literature starts around 2008 or so. I can't give you references since it's not my subfield but it is done. (Daron Acemoglu is one name, but he works on many topics. Maybe Douglas Gale has also worked on this.) Models of contagion in networks among banks are an example.
There is also literature in international trade on trade networks (sourcing components for a final product, for example). Here I don't have names for you.
There is also Matt Jackson at Stanford who has worked on many, many topics in networks. On the empirics (which are very challenging) you may want to look up Bryan Graham and coauthors.
From my limited exposure the work on financial crises and trade, it doesn't seem that interesting but it does exist. The empirical work on networks exposes a lot of challenges. Graham (IIRC) has a recent survey in the Handbook of Econometrics if you'd like to learn more.
Matt will get the Nobel for it.
Entirely possible and very hard to predict but... I am not that convinced a lot of the work is impactful in the broader profession? Even within theory is this still a hot topic? I feel like this is something everyone was excited about around 2010++. [Not my subfield so you can tell me the answer is yes if you are an expert.]
And ... is networks worth a solo prize? If not, I am not sure who else he would share it with? Or how they would shoehorn this into a "Networks and..." prize. [If you are a micro theorist please feel free to correct me.]
I think it is still struggling to develop the necessary impact, although the work certainly is of that quality.
And I say that as an admirer and student of Matt, having also worked on these topics within that circle.
What would be a single theoretical result to highlight? No doubt, many beautiful results exist, say, existence of pairwise stable networks, equality of Nash equilibria and centrality, impact of complementarity or substitutability on multiple equilibria, information transmission etc. for an economist or mathematician that's exciting stuff.
But to get a Nobel you need either empirical impact and you need a soundbite (I think). A single idea - or a few of them - that encompass what it is about.
Network results rely on structure. That's what it is about. Everything else can be modeled by more parsimonious game theoretic models (and has been).
But what does structure offer that has not been explored by physics and sociology? Structural holes? Complexity? Phase transitions? That all predates econ.
To be fair, Matt Jackson has probably done as much as is possible in achieving so many outcomes - and in so many areas. Like showing that canonical network models do not fit all empirical facts - e.g. preferential attachment leads to fat tails but never to positive assortativity and so real life networks also need second order discovery. Or showing stuff about dominant groups in information cascades etc. etc.
It's just so much... but there's just no one thing. And it is all beautifully theoretical.
My prediction: The Nobel will come via empirical applications. Here, network structure matters a ton. Multiple equilibria, networked externalities and the associated identification issues are omnipresent and a part of life.
At some point, we will learn so much with these techniques they will give a Nobel, and then they also have to give it to Matthew.
Interesting, I do not understand what do you mean by "equality of Nash equilibria and centrality", please could you elaborate or provide a source?
Good questions and good points... something that bothers me is that networks are treated into micro when they actually relate to meso... connecting micro and macro.
Impossible to discern who would share the prize. Also, his network science is childish compared to nowadays econophysics multiplex approaches, but this was never a deterrent for the comittee.
Finding the ideal meta-structure is not the core problem. It's getting every company on earth to report the data remotely accurately and in a timely way, then disseminating and analyzing it.
I actually think that we need government to enforce this data collection and it needs to take advantage of some decentralized systems for it to be workable. Primarily because we need "hard" (usable) data about resources, wealth (inequality) and crops etc. in order to have a realistic (and indisputable) view of what's happening. Combining that type of decentralized megastream with advanced cryptocurrency smart contracts could change economics from being a cult to a useful science.
My question is more focused on what the ontology would be for different purposes?
What are the node types, properties of the nodes, and the edge types to best model a catalyst (like the Ukraine war).
Geopolitical hierarchies and other relationships which are related to ethnic power struggles.
Sounds like you want to build a fully-working and -encapsulated simulation of the universe ;)
Even modeling a single, non-trivial business would probably be exceptionally difficult
If you make the nodes only "public companies" and the edges "have a business relationship" it is not an exceptionally difficult, would be pretty trivial.
The question is what is an appropriate ontology to have some basic sense of how a catalyst can affect a company.
If a catalyst affects nodes, how many steps until it hits the "company node" in question.
There are ~6k public companies [0] just in the US
OP then lists at least 6 prominent factors to add onto "merely" your overly-simplified b2b relationship :)
Simulating the entire publicly-traded economy is - effectively - on the order of simulating the the universe :)
--------
[0] https://www.benzinga.com/news/20/10/18026067/the-number-of-c...
You need to simulate how companies react to stuff. Might be possible using some statistic models though?
I mapped my country's businesses and their owners relationships. Some very interesting things popped up. A private yellow/fake news company with THOUSANDS of owners/shareholders, lots of people with few hundred companies. company "rings" etc. Pretty fun stuff :)
Very cool! Do you mind posting what ontology you used for the graph? It is difficult to figure out what is a "node" or what is a "node property". Is "energy demand" of a company a property of the company node, or is that a node itself that connects to energy supply companies?
My ontology is very simple, since I only care about owners/shareholders
2 types of nodes
I can start adding other types of relationships and properties, but that was not my goal :)I did a smaller subset of what you mentioned as part in my first attempt at a PhD (since abandoned) around 15 years. What I did was to model relationships between banks with edges denoting interbank loans. Then I computed the eigenvector centrality (think PageRank) to characterize the "risk" taken on by a bank. This was later expanded to include things like mortgages, prime and subprime loans.
The idea was that you could later run simulations and what-if scenarios. Lots of agent-based modeling happening too once we had the structure up and running.
(this was done around the Lehman Brothers period, and there was a lot of interest in these kinds of works in Complexity Sciences).
Why did you abandon this? I'm assuming the modeling assumptions were too simplistic.
I did think up some ideas one evening long ago and wrote down a rough note, but I don't think it would have much use for serious predictions (except as a fun toy to play around with to generate ideas)
A mostly un-edited, totally unpolished, and probably erroneous - don't judge too hard :) - version here:
- All goals are to some extent intermediate - a means to achieving a further goal - directed but not acyclic
- Some goals are largely measured by how useful they are achieving others, exemplified by stocks and tokens
- Node values are the measurement of the goal (in what?) and the edge values are the percentage split (like a Sankey diagram) but inclusive of factors less than 0 or greater than 1 (i.e any value) to be added to the value in the node, like y = y + (xf)
- (The x-f relationship could also be exponential - (xf^n) is a better equation)
- Maybe, by measuring the values of x and y empirically over time, we can try to calculate f. f has units to balance out the units of x and y, so no problem with incompatible units
- What are the nodes? Every damn thing that can be measured - prices of everything sold on the market, population statistics like literacy rates, time spent on Khan Academy, anything that can be quantitatively measured (quality of the measurement doesn't matter as each f value is completely independent of other f values)
- And we have a tech tree! You can choose the measurements that you want to optimize for and use it to prioritize your resources towards progress. Can also be used to intelligently guess at the inputs and outputs of progress in a specific goal
- Better for quantifying the current economy and scientifically deploying investment for the near future. The long term is obviously unpredictable (think https://twitter.com/robert_zubrin/status/1278681124944793611), however can be used to analyze changes during previous paradigm shifts with historical data
- This is 99% dependent upon price signals (which I believe will be almost all of the useful data)
This is very helpful thank you :)
Having a general idea of what can be affected is the goal, not an accurate prediction of the future universe
How can the Ukraine war affect Subway (restaurant)
Ukraine war (catalyst) -> Key exports of Ukraine -> Grain -> Acme Grain To Flour Co -> Subway
Supply/price of flour may change for Subway.
Node Types: - Country/Region (Ukraine) - Raw Material (grain) - Company (subway / acme grain to flour co)
Relationships - Export - Input to - Supplier to
We're working on something very similar in the CoyPu Research Project. You can find our preliminary OWL ontology here (please note it's very much work-in-progress): https://schema.coypu.org/global
A release of a sanitized public version of the materialized knowledge graph (sans internal company data) is planned for the near future. If you'd like to work with us please get in touch!
Do you see a need to have a good ontology of describing generalized supply chains? I see a few classes that cover this but am not sure if it is a focus.
This is great, you will be hearing from me :)
Most of the "knowledge graph" world takes an "outside of time" viewpoint. When you get into more dynamic situations like you are talking about (run simulations) you are getting beyond state of the art.
A basic model, or general ontology is the scope. Very much agree dynamic situations and finding equilibrium is definitely a non-trivial question.
https://news.ycombinator.com/item?id=33712337
But why? We can trivially modify any point in a graph at any time.
It’s certainly possible but you don’t have a lot of help in doing so.
OWL from the RDF world has quite a few reason it hasn’t gotten traction but one is that it doesn’t have answers for many problems of ‘agentic’ reasoning which would be useful to agents that are doing things in the world.
There are problems with the management of a database in practice. A document database lets you update a document at a time, a relational database a row at a time. If you are adding edges and nodes to a graph there needs to be a transaction facility so that you never see inconsistent states. Deleting a ‘record’ is not just a matter of deleting a ‘node’ or an ‘edge’ but may involve deleting some set of nodes and edges which graph databases don’t have in their model.
In the case of OWL I think you need a transaction mechanism to make the data and ontology checking abilities of the standard useful. In a normal OWL database you put junk in and now you have an invalid database, error messages that don’t make any sense, and no systematic answer to fix it. Really in that case it should revert the transaction or maybe let you work in an invalid branch but never be able to commit an invalid branch. So far as I know most OWL products today have no real answer for this problem but it seems to me they could.
Generally in common sense reasoning there are many issues like ‘what was the state of the world on Nov 4: 1975?’ Or ‘what does John believe about X?’ or ‘it is necessarily true that…’ or ‘what would happen if?’ that go beyond first order logic.
Many knowledge graphs build structures that let you query who was president at a particular time because there are records that say ‘X was president between date A and date B’ but this only works when it is spelled out explicitly, you can’t ask questions about the world as a whole at a certain time.
Some of these problems could be solved by the use of persistent data structures, that is, fork the world when you put in facts so you can’t break the world, fork it to do a simulation, etc.
This idea was also documented by Mindey on Infinity Family.
It was called network of functions.
https://o2oo.li/method/222/#1629594291
We at mappes.io are doing something similar in industrial domain. Our knowledge graph has few layers
Products (Raw Material <> Application use) Products to company (Supplier <> Buyer) People connected via companies and products
We are already seeing benefits of this in being able to easily discover new connections across products and companies. Our focus is right now on few verticals in manufacturing sector and hope to expand to wider manufacturing space at some point.
Try the work of Wynne Godley and Marc Lavoie, who use stock-flow consistent models to model the economy. See their book: Monetary Economics: An Integrated Approach to Credit, Money, Income, Production and Wealth [1]
[1]https://www.amazon.com/Monetary-Economics-Integrated-Approac...
This is interesting thank you. The approach seems to be "money/cash centered" from what I am gleaming.
I am wondering what a "company-centered" approach ontology would be.
When politicians are setting national budgets – arguing for a billion here, a billion there – I always wonder what model they use. Is it a spreadsheet? What are they using?
Take a look at Stock-Flow consistent models which incorporate some of what you are looking for.
Agent based economic models were a really hot idea 15-20 years ago. They had interesting properties but to my knowledge nobody ever could calibrate them to generate real-world testable predictions.
If you really want to get exotic peek at what the Cybersyn people were trying to achieve.
Haven't heard of the "stock-flow consistent model" before that is helpful I appreciate it.
This thread may be of interest as somewhat related, with modeling of the energy flow at global scale. https://news.ycombinator.com/item?id=31966435
Basically, a digital twin of the world economy. There are conspiracy theorists who think that this exists and captures all of the major firms and simulates the rest out for small businesses.
Interesting way to put it, "digital twin of the world economy".
What would be the simplest ontology definition to get a rudimentary idea of what's going on?
Hackernnews needs more diversity:p This problem is studied in industrial ecology literature and complex networks related research.
I'm working on this.DM me if you're interested
Are you aware that HN has no DM function and your profile does not list contact info?
saw an article mapped of hundreds and hundreds of thousands of businesses and map them to other companies to see what companies actually ruled the world. There were like 6 or 8 companies that were at the center of all other companies.
I forget exactly the details but it was a cool article.
This could be done. What would be the source of data? Would it be updated in real-time?
LinkedIn?
Improbable might be working on this.
Welcome to complexity economics.
Does complexity economics have a working ontology to model how companies/regulations/supply chains interact with each other?
It has a data driven approach that does not sum up into an ontology, still it flies.