Now that I think about it the matrices in front of and after the adjacency matrix could be generalized as the upstream weights and downstream weights respectively, as a function of anything not just labor content
@casperadmin I guess the difference would be if you're in a country that is expanding its oil production to meet the demands of a growing population whilst oil is provided as an input to many different industries an interruptions in the production of drilling equipment which only provides inputs to say oil and mining percolates to oil production and every other sector dependent on it.
another example would be silicon, silicon itself provides inputs to glass, semiconductors, other chemical products etc. but those three downstream industries provide inputs to a lot more than three industries themselves.
Ah I see your point. So rather than simply looking at the sector which serves as the biggest share of imports, you'd want to look at each sector with a compiled output that includes industries which use as inputs the outputs of industries where the given industry was an input. In other words:
- Industry A provides inputs for industries BCD
- each of those industries provide inputs for industries EFG
- hence industry A is essential for all those other industries, even if it isn't counted as a direct supplier for EFG.
I think the problem you'll run into is that since input-output tables that exist today are on the level of industries that include a wide array of products, it's possible for only certain productive processes/firms in that industry to play the role of indirect inputs.
Keeping in mind such limitations, I think this is still place to begin research. Once you know which industries are most central, then you can zoom in with further research.
I wish I could be more help with the math and implementation in python, I'm much more familiar with R.
If I was approaching the problem from scratch, I would want to compare input-output tables between years, create a vector for each final demand industry that contains its direct and indirect inputs, look at the change between the years and run regressions on each input to see which one contributes to change in output the most, then weight each final demand sector by size.
If I was approaching the problem from scratch, I would want to compare input-output tables between years, create a vector for each final demand industry that contains its direct and indirect inputs, look at the change between the years and run regressions on each input to see which one contributes to change in output the most, then weight each final demand sector by size.
This is good to know, if my generalization holds I should be able to incorporate any weighting scheme
I think the problem you'll run into is that since input-output tables that exist today are on the level of industries that include a wide array of products, it's possible for only certain productive processes/firms in that industry to play the role of indirect inputs.
If this plays a significant role it will likely be the case a bunch of sectors come up with centrality scores very close to one another, so this is good to know.
This also highlights the necessity of developing methods to indirectly measure I/O tables of which I have no clue where to look lol
If this plays a significant role it will likely be the case a bunch of sectors come up with centrality scores very close to one another, so this is good to know.
This also highlights the necessity of developing methods to indirectly measure I/O tables of which I have no clue where to look lol
If you didn't have access to I/O tables but had access to income tables by industry, you could check correlations between industries and organize them by capital/labor intensity. Alternatively, you could go the opposite direction, from the smaller picture to the bigger picture. Look at financial statements from leading companies in an industry and figure out what their major inputs are that way. Oftentimes companies will actually list their vulnerable bottlenecks in their "risks" section, as they're obligated by law to give such warnings to investors if they're aware of them.
@casperadmin This is definitely the next step after the theory is fleshed out; accurate data is crucial for the application to be as dynamic as possible, the ideal would be enabling a kind of fine-grain persistent action that can be as effective as a general strike.
@madredalchemist I'm unfortunately not nearly as well-versed in computer science or cybernetics. But this discussion with regards to identifying the weakest links in the production chain reminds me of Forces of Labor by Beverly Silver. Specifically, her discussion of why the most militant sections of the nineteenth century European working class (textile workers) were less successful than their contemporary railroad workers or longshoremen. An answer I found persuasive was that textile capitalists learned to reserve small amounts of semi-finished goods so that when textile workers went on strike, the capitalists could potentially outlast them. The instances in which the spinners won their strikes were when they were joined by sympathy strikers in either other parts of the production process or from other industries. If they weren't buttressed by community support, textile workers were usually outlasted by the capitalists.
However, as Silver later demonstrates, does not work throughout all real commodity industries. For instance, the twentieth-century auto workers were able to inflict far more damage because semi-finished cars took up far more space to store than semi-finished pounds of yarn. This meant the capitalists had fewer options in simply "waiting out" the strike.
I don't think this bears directly on your analysis, and my knowledge of cybernetics limits me from developing a meaningful critique of your work other than that this seems to be very promising stuff. I bring up this history because a) Silver's work may be relevant to what you're figuring out re: developing methods of figuring out the weakest links in the value chain and b) the physical properties of real commodities can limit the options capitalists have in dealing with striking workers; I don't know how point B can be factored into your calculations, but as a qualitative point it may be of use to your research.
@jules1214 This is good to know but I thought commodity reserves have gone extinct with just-in-time production? If the pandemic was any indication it seems like firms aren't stockpiling anything anymore.
@madredalchemist I think with the pandemic-induced supply shock a lot of consumer goods firms have been moving away from JIT production. At least from what I remember reading the Financial Times. Especially since stretching your supply lines so thinly meant creating a large number of choke holds for logistical workers (like the longshoremen at the Port of Los Angeles) to exploit. But I yield to the superior knowledge of any logistics specialists who are more in the know about it than I am.
@jules1214 Regardless if my modeling is valid I can weigh the centrality by anything so I could weigh the centrality against reserve levels of any industry
@casperadmin This is definitely the next step after the theory is fleshed out; accurate data is crucial for the application to be as dynamic as possible, the ideal would be enabling a kind of fine-grain persistent action that can be as effective as a general strike.
It occurred to me in the shower earlier that if you were looking at correlations in income, that during a supply chain crisis you would actually see the correlation be negative between the industry providing the input that's in shortage and the final demand industry, as the shortage would cause a distributional shock with the higher prices for the good.
If people could share the thread with any math/cs folks they know it'd help to get some extra eyes on it to validate the model.
Is Ian Wright or Tex (from https://www.youtube.com/c/TexTalksSometimes) on this forum?
@madredalchemist I think with the pandemic-induced supply shock a lot of consumer goods firms have been moving away from JIT production. At least from what I remember reading the Financial Times. Especially since stretching your supply lines so thinly meant creating a large number of choke holds for logistical workers (like the longshoremen at the Port of Los Angeles) to exploit. But I yield to the superior knowledge of any logistics specialists who are more in the know about it than I am.
Some manufacturers keep stockpiles some don't. Keeping a stockpile can be very useful sometimes because in a shortage you can have immense control over the market, but it can also be very expensive. As we move away from just in time production I'm sure more will also embrace some level of stockpiling.
@casperadmin How far back can you go with the this data? i.e would data from the pandemic supply shocks still be relevant now, 1, 2, 3, 5 years out etc?
@madredalchemist it depends on the shock. In terms of the initial pandemic shocks we are past that, and even with the tight labor market labor isn't getting any extra real income. But we can see pretty easily how the war in Ukraine has created higher revenue and profits for the oil and natural gas industry. But obviously, the shock will persist for however long the shortages do, which is dependent on material facts on the ground.