Skip Navigation
39 comments

  • Can you explain how you are calculating the wages paid out to workers? In words you state:

    c. It is assumed that consumers are not going into debt, which means the total money they spend on net consumption comes from the wages (or dividends) paid by industry. Wages are paid by each industry proportionally to their gross labor usage

    And in the code you have:

    ` %Industry to consumer sales (assuming consumers save/lose nothing)

     
            S = o.*P; %Sales to consumers

    Y = sum(S); %Total consumer spending = wages paid out

    W = L.*Y; %Wages paid out vector by industry

    I have a couple of questions about this, but I still start with the calculation of Y (the answer may clarify my remaining questions).

    You start with S = o.*P; %Sales to consumers where you are performing an element-wise multiplication of a n x 1 vector of net-products by a n x mag array of prices.

    Working under the assumption that the net product is the net consumption of workers, i.e. n = c (also, don’t worry about using a standard notation - there isn’t really much of one. I am using a mix of Ian Wright’s, Pasinetti’s and my own

    ) the line S = o.*P; is calculating the below, correct?

    Each of these elements, si = pi ci is the part of sector i’s revenue which it receives from its sales to consumers. (It isn’t the total revenue of the sector, though. That is given earlier by Ri = pi qi).

    Then you are summing along each column (you mentioned that MatLab’s sum function sum’s matrices down the column). For a single instantiation of a price vector, this is results in

    Y = p1c1 + p2c2 + … = p c

    which gives us a global wage bill, i.e. the total expenditure of the working class for the economy. This is an aggregated quantity over the entire economy.

    Y is a row-vector, though, in your code since you are testing multiple prices - mag of them in fact. if I have the correct understanding, your Y vector is

    Y = [Y(1) Y(2) … Y(mag) ]

    where I am using superscripts to designate distinct price instantiations.

    But I think I must be misunderstanding something, because if the above is true, then I am not certain how the line

    W = L.*Y; %Wages paid out vector by industry

    Works out because you couldn’t element-wise multiply a n x 1 array L with a 1 x mag array Y.

    My understanding is that your L vector is a n x 1 array of normalized labor-times for each sector.

    L = O.*l;

    L = L/sum(L); %Normalising gross labor use

    So I need your help in understanding the calculation here.

    I have more questions, but I will save them for later. Thanks for taking the time to discuss this!

    • Working under the assumption that the net product is the net consumption of workers, i.e. n = c (also, don’t worry about using a standard notation - there isn’t really much of one. I am using a mix of Ian Wright’s, Pasinetti’s and my own ) the line S = o.*P; is calculating the below, correct?

      Yup, 100% correct

      Works out because you couldn’t element-wise multiply a n x 1 array L with a 1 x mag array Y.

      This is indeed a valid operation in Matlab, also equivalent to L*Y (I could have clearer here)

      Thanks for taking the time to discuss this!

      No problem

      • Okay great! This answers my next questions

        So you calculate an average wage-rate (price per labor hour)

        w=pc/L

        and pay out to each worker sector i their eage using this wage-rate

        wLi

        Now we have a price equation that we can represent as

        p = wl + pA + π

        where the profit of each sector i I'm defining as Profiti=πiqi. I.e. πi is a profit per gross output of a sector.

        The above is in a style that is close (but not completelt identical) to a Sraffa-Pasinetti framework. But setting it up that way can allow you to compare with the Sraffian framework that you may find in the literature. In this framework, it is assumed that due to competition of capitals, an equilization of profit rates emerges. A single profit rate r emerges. So π becomes written in terms of the rate of profit, π= r pA, or it is sometimes defined as r (w l + pA) which is more in line with how the classical political economists such as Marx defined the rate of profit.

        Also, if we assume the capitalists take their profits and spend it all on commodities such as luxury items, then we will need to introduce a capitalist consumption vector cK and add it to the worker consumption so it is included in the net product. This will increase the gross product, and also increase the labor that workers must provide (i.e. now workers must perform surplus labor to create commodities for capitalists). Ian Wright's contribution is finding a measure of value which incorpotes that surplus labor and is proportional to the price. It is a measure of total labor embodied in a unit of workers' consumed product.

        For the model in this post, there is no assumption of equal profit rates, no structual mechansim for enforcing equalizafion of r, so the profit is simply a residual.

        If ever interested more in any of the above I'd always be down to talk. I find it fun!

        Here are some extra tidbits about the equation. You may or may not find this worthwhile for the purposes of the simulation.

        Again, the equation you have is

        p = wl + pA + π

        We can rewrite it as

        p = wv + π(I-A)-1

        This tells you the difference between the price and the standard labor value in terms of your residual.

        Another alternative way to view the original equation is purely in terms of quantity flows, if we note that w=pc/L, then the term wl can be written as

        wl = p [(cl) (1/L)]

        The term in brackets, [(cl) (1/L)], is actually a consumption matrix, C(W), for the working class. Recall c is a column vector while l is a row vector. It is like the workers' consumption analog to the industrial sectors' input output matrix A.

        Element i,j of A tells us how much of product i the sector j requires to produce one unit of gross output in j.

        Similarly, element i,j of C(W) tells us how much of product i workers in sector j consume per unit of gross output in sector j.

        Some interesting properties of C(W)

        C(W)q = c(W) (for the workers)

        pC(W) = wl

        So if C(W) is known data, or constructed by a model, you can write the equation

        p = wl + pA + π

        as

        p = pC(W) + pA + π

        which can be rewritten as

        p = π (I - A(+))-1

        where A(+) = A + C(W)

  • Thanks for your comments and clarifications! I'll read them and get back to you. The use of the 1000 price vectors within each iteration, and the explanation of some MatLab code helps

  • I have some comments, questions, and possible suggestions. I think this is really great, and simulations are incredibly fun and I love digging into them. Some of my questions are because I know Python but not MatLab, so I may need to check I understand what your code is doing. Other comments may be due to differences in how quantities are calculated (particularly how the labor-value and the costs were calculated), and your final comparison of aggregated quantities instead of sector quantities. Keep in mind that I don’t mean any comment to sound aggressive, and it is possible that I am misunderstanding the code or the concept that the calculation represents.

    Because I don’t know MatLab well, I’m going to go through the sections of the code and we can confirm my understanding of each part.

    1.) You make N economic simulations. For each iterative economic simulation, i, you:

    2.) First, generate a random net product vector n for the global economy (you call it o in the simulation).

    3.) Then you randomly generate an input output table A.

    • Warning: not every randomly generated input output table is productive, i.e. there is no guarantee that the inverse of I-A exists, or that if it does exist then you will have an economically feasible gross product. The Hawkins-Simons condition (which is an economic application of the Perron-Frobenius theorem) gives us the mathematical conditions for ensuring that an input output matrix A is economically feasible.
      • Essentially for A to be productive, (I-A)-1 must exist and (I-A)-1 n must result in a non-negative vector (you can’t have a requirement of negative gross production) which means (I-A)-1 must also be non-negative. -- To confirm that A is productive. You can check if the largest eigenvalue of A. If it is a positive value that is less than one then by the Hawkins-Simon theorem it will be productive. -- Alternatively, you could just randomly generate the input output matrix A and then confirm that (I-A)-1 exists and is non-negative. If not, regenerate A and test again. -- I am also sure that MatLab would give an error if the matrix inverse does not exist, but it won’t give an error if the resulting gross product is not non-negative.

    4.) Then you calculate the gross product vector q, or as you call it O. This is calculated via q = (I-A)-1 n. It took me a while to realize that “\” is MatLab’s way of doing a matrix inverse followed by a multiplication. So A \ b is MatLab’s way of calculating A-1b, correct?

    5.) You calculate the gross labor use as L=l q. This is element wise multiplication, or equivalently, the dot product. I wasn’t sure why you normalized the net product and the gross labor, though.

    • I added in the explict checks. It turns out, a huge number of the matrices being produced were non-productive. Instead of trying to keep generating matrices, I made a different fix (which makes the technical matrices more realistic, so win-win)

      I made it so that the average number of entries in each row of the technical matrix is (2*n)^0.5

      This means that as the economy grows larger, the matrices grow sparser. This makes productive matrices much more likely (at which point, I just have a check which makes it so that non-productive matrices are regenerated).

      Curiously, this change doesn't have that big of an effect on the outcome. I've verified. The model simply handles negative net production and treats it like purchasing commodities from the external market (so something like imports). Still, I have removed it for now.

    • 6.) Calculating the LTV “prices” is the part that confuses me the most and I have a few comments and questions on

      • You calculate this by first calculating (I-A)-1, and then perform an element-wise multiplication (this is what “.*” does in MatLab, correct?) by the labor coefficient vector l. Then you sum this together.
      • I am confused by this calculation as the standard equation for calculating labor values is v = v A + l = l (I-A)-1. Note that l is a 1 x n row vector and so v is also a row (or left) vector. Alternatively, you could calculate with column vectors and the transpose of the inverse, i.e. v = ((I-A)-1)T l.

      This gives you the amount of labor required in a vertically integrated subsystem to produce one unit net product. This is using Sraffa and Pasinetti’s work to give a more concrete theoretical understanding of what a labor-value is. It is like a total labor input per unit net product. By that definition, though, it isn’t meaningful to sum up these elements as they don’t have the same units. You would first have to multiply v by some commodity-quantity, typically the net product. Note that v n = L, this provides an alternative way to “dividing up” the economy’s social labor.

      • This is similar to how you couldn’t actually meaningful sum up the elements of a price vector, p. A price has the units of money per unit commodity, and each commodity would hence have a different price unit. You would first have to multiply each price by the quantity of commodities that one is purchasing in order to convert it to a common unit (money) and then you could add it up. The same logic works for the standard definition of the labor value vector.
      • Another reason to not sum the value vector is that it would be useful to compare each element of the value vector, i.e. each commodity’s labor value, to the emergent price in the market. But, this would require you to add a mechanism where each firm can adjust prices and I don’t believe that has been added into the model yet. At the moment you are comparing aggregate quantities, i.e. the sum of prices with the sum of values-per-commodity (which I am not confident is meaningful in your present calculation), but an improvement could be to compare each sector’s emergent price (the price required for reproduction) with the sector’s labor value. In another post I have some papers where you can read how Ian Wright’s simulations handle the price adjustment and the reallocation of labor. This can be one possible future direction to head toward. I.e. you can inspect the ratio of pi / vi for each sector i.
        1. uh, I haven't really read staffa or anything, but my approach for computing ltv prices was

        Assume 1 unit of net output for a commodity

        Calculate how much gross production would required for each sector in that case

        Element wise multiply this with direct labor use to get needed labor from each sector

        Add up all of the labors from all the sectors.

        In formal terms, for sector 1

        (I-A)^-1 * [1;0] = g = Gross product for 1 unit of sector 1

        Then

        sum(g.*l)

        The "(eye(n)-A)\eye(n))" computes "g" for every sector at once (the output is [g1 g2 g3 g4 ...])

        I believe this is equivalent to the equation you have also provided, except your equation involves fewer steps.

        Also, just as a context thing, matlab, for some reason sums matrices down columns by default. So the output the ltv prices equation is a row vector of the summed labors (which I transpose using the apostrophe symbol ')

        Indeed I am pretty sure that my code actually is using prices and values for each sector separately, otherwise the code should be giving me an error regarding the dimensionality of the code.

      • 7.) Then you have a vector of randomized prices. This is an n x 1000 table, though. I am confused why there is an extra 1000 here, when the loop itself will create 1000 instances of the economy. I may be misunderstanding the intention of this step so any clarification can help.

        8.) You then calculate the costs, the sales to consumers, the total wages paid out, and the net income (I am also confused why the net income is normalized)? Some of the cost calculations confuse me, so I will discuss my understanding of calculating cost and we can also check if we are converging on our understanding: Sector i will spend the following amount of money on their means of production per unit product:

        unit costi = p1 a_1,i + p2a_2,i + …

        This is the amount of money that sector i must shell out to all other sectors in order to built a unit quantity. You can think of it as the row vector of prices multiplied by the i-th column in the input-output matrix A.

        This unit cost for each sector can be expressed via matrix algebra as

        unit cost = p A where p is a n x 1 row vector of prices

        This gives you a row vector of each sector’s unit cost.

        Alternatively, you can calculate this as unit cost = AT p where p is a column vector. I just prefer to use column vectors for physical quantity vectors, and row vectors for prices, labor coefficients, and values. It makes the math easier to write, and you avoid transposing. It also hints at a duality present in the system.

        Since this is a unit cost, we can calculate the total cost of production for sector i as

        costi = unit costi x qi = (p1 a1,i + p2a2,i + …) qi

        You can do this for all sectors by first calculating the unit cost vector above, so you can do p A using matrix multiplication, then you can convert these unit costs to real costs by element-wise multiplication with q. This will give you each sector’s individual costs for means of production.

        If you wanted aggregate costs for the whole economy though, then you can do the matrix multiplication of p A q, but I would suggest moving from an aggregate to a sector-specific model so you can test the differences between sector prices and sector labor values. I think that would be very interesting!

        The money that each sector receives would then be the element-wise multiplication of pi qi. If you wanted an aggregate quantity across all sectors, this would be the p q - you have already calculated this as R.

        Then you also need to include the wage payments for each sector. Each sector pays there workers w li qi = w Li. In the aggregate if workers are spending all of their wage and they are the only consumers of the net product, then w L = p n. But if you wanted to find how much each sector was spending for their wages, then you have to disaggregate this. You could Introduce the wage as a new parameter. Then the total wage spend by each sector is w li qi = w Li

        Divide the working class into sectors, just as the industry is divided, and note that the i-th working class consumes a proportion of Li/L n of the net product. Then, by conservation of the worker’s wages with their expenditure, note that wage spent by each sector is then pi ni Li/L.

        With that, you can then find the net income as you have. And perhaps this is what you’ve done, but just in an aggregate way. As I mentioned, MatLab code is a little difficult for me to understand. And also, I think disaggregating this can be a good next step!

      1. correct
      2. correct
      3. I was under the impression that since matlab's "rand" generates values between 0 and 1, all of the technical matrices should be automatically productive, but you are correct. I should add explicit checks
      4. yes
      5. I normalise the net product so that different economies are more comparable to one another. I am interested in the ratio of output between 2 industries, and not the scale of overall production

      I normalise the gross labor use because that variable should actually be "share of workforce employed". The name is a relic from my previous attempt at the code where I had tried to incorporate the population into the sim. I seem to have forgotten to change the name.

      Matlab's elementwise multiplication is not a dot product

    • I have some comments, questions, and possible suggestions.

      Exactly what I'm looking for :)

      I'm not exactly a mathematician, so I had intended the simulation to more so be a learning excercise. Feedback is appreciated. I'm going to take a look at your comments once I am done sleeping 😴

  • This is cool. Have you read any of the people who have done work on LTV and input-output analysis (or econophysics) type stuff? I feel like you'd dig it, it's very data based.

    Like he sucks as a person, but I feel like Paul Cockshott would take a look at this if you sent it to him.

    • I learnt it from the guy in the first place. But yes, that's a great idea. I'll see if I can contact him.

      • That makes sense. There is also the econophysics discord. Nicolas Villarreal may also be interested, he does some simulation type stuff for profit rates.

        You may also like some of these. I've only read one or two, but have been collecting them.

        Cheng, Han, and Minqi Li. “Do Labor Values Explain Chinese Prices? Evidence from China’s Input-Output Tables, 1990–2012.” Review of Radical Political Economics, vol. 52, no. 1, Mar. 2020, pp. 115–36, https://doi.org/10.1177/0486613419849674.

        Diquattro, Arthur. “The Labor Theory of Value and Simple Commodity Production.” Science & Society, vol. 71, no. 4, 2007, pp. 455–83, https://www.jstor.org/stable/40404443.

        Flamant, Christian. “The Labor Theory of Value and the Problem of Joint Production : The Failure of Sraffa’s Theory and Morishima’s Misconception.” World Review of Political Economy, vol. 14, Apr. 2023, pp. 63–98, https://doi.org/10.13169/worlrevipoliecon.14.1.0063.

        Lee, Chai-on. “Marx’s Labour Theory of Value Revisited.” Cambridge Journal of Economics, vol. 17, no. 4, 1993, pp. 463–78, http://www.jstor.org/stable/23599796.

        Li, Bangxi. “Marx’s Labor Theory of Value and Its Implications for Structural Problems in China’s Economy.” Economic and Political Studies, vol. 2, no. 2, 2014, pp. 139–50, https://doi.org/10.1080/20954816.2014.11673848.

        Nikolaos, Chatzarakis, et al. “Does the Labour Theory of Value Explain Economic Growth? A Modern Classical View.” SSRN Electronic Journal, 2022, https://doi.org/10.2139/ssrn.4133135.

        Slack, Gregory. “Marx’s Argument for the Labor Theory of Value.” Review of Radical Political Economics, vol. 53, no. 1, Mar. 2021, pp. 143–56, https://doi.org/10.1177/0486613419895275.

        Tsoulfidis, Lefteris, and Dimitris Paitaridis. “Chapter 8 On the Labor Theory of Value: Statistical Artefacts or Regularities?” Research in Political Economy, edited by Paul Zarembka, vol. 25, Emerald Group Publishing Limited, 2009, pp. 209–32, https://doi.org/10.1108/S0161-7230(2009)0000025011.

        Zachariah, Dave. Labour Value and Equalisation of Profit Rates: A Multi-Country Study.

        All are likely on libgen, and the last one I think you can find on his website.

  • Hey this is awesome I've done some rebuilding of Ian Wright's simulations in Python. I'd love to share ideas and approaches and compare. He has his own idea as to how the transformation problem gets solved and the simulations built seem to support it (at least that it's consistent)

    There are also two approaches of these simulations

    1.) Is a macro approach. You simulate each sector as it's own node or "aggregate firm" and let the prices adjust by some macro economic function. You could test out different functions.

    2.) There is the micro approach. Less has been done here, but it is definitely the more fun approach. You model each firm and/or individual (each agent) and let them set their prices, activity levels (how much they produce), etc.

    In both simulations, you let the agents, whether it be a aggregated sector representing many firms in the macro approach or a individual firm in the micro approach, interact and change their own states as well depending on how much money they make. You need some mechanism of "decision making". Either dynamical equations, generative or stochastic actions, or more advanced things like Q learning agents (Ive used decentralized deep Q learning agents but it gets computationally intensive and isn't scalable to a large micro model. And many "dumb" agents are likely superior and sufficient to intensive "smart" agents for micro models)

    The two types of models have different types of economic systems they are simulating, but in each simulation the appropriate activity levels are found (the agents essentially solve q = (I-A)^-1 for the global economy) and the prices correlate to labor values. In the macro models Wright shows the prices converging to the super vertically integrated labor values which he claims are the appropriate measure of value which correspond with price.

    Essentially, the indirect value that goes into producing luxury commodities, which is what the profits in the second term of p=w l + r p A + p A get realized as, should be included in a measurement of value to be commensurate with price. If one wants to discuss the technical requirements of production then the standard labor values v are to be used. If one wants to know the social requirements of production due to exploitation and have a measure of value that also relates to the price structure, then you can use Wright's non-standard value measure.

    Seeing if you can confirm or reject that work would be an interesting direction to take this in

39 comments