Modelling the Experimentally Organized Economy (II)

The Stockholm School (theoretical) point of distinguishing between ex ante plans and ex post outcomes, and how agents learn from feed backs of mistaken anticipations that affect their behavior and influence macro, are distinguishing features of the MOSES model economy (Eliasson, 2015 ), as is market competition driven by entrepreneurial entry, that sequentially both forces agents to perform, or exit, and inform them through price signaling how to allocate their resources. In that general market perspective individual decisions (in firms) can appropriately be likened with the design of economic experiments to be tested in markets (Eliasson, 1987b; Eliasson, 1991a). The MOSES model thus integrates bounded rationality characteristics of agents (Simon, 1955) with the push of entrepreneurial entry (Schumpeter, 1911) that creates a collective competitive force that compells incumbent agents to innovate preemptively, not to be competed out of the market (Eliasson, 1995; Eliasson, 1996). All this dynamic occurs within a complete micro to macro and stock flow consistent Keynesian demand accounting scheme, and a Leontief input output production structure.¹ By taking the macro analysis down to the micro market level a theoretical link to ; Marshall (1890) and Marshall (1919)'s concept of industrial districts populated by micro agents is furthermore established. With the entire model economy being self-coordinated by the invisible hand of markets, the MOSES model takes on not only distinctly Adam Smithian qualities, but the dynamics of market selection also creates the positive network productivity externalities that Marshall verbally formulated,² but that later became the cornerstone of so-called New Growth theory (Romer, 1986). In the MOSES model, however, network externalities are explicitly created by market induced reallocations of resources among heterogeneous agents, that change the organization of markets (see Section 4.4.4). In fact, one novelty of the MOSES design is that selection of technological innovations driven by entrepreneurial competition, most prominently through the entry and exit of business agents through a Schumpeterian type of creative destruction (see Table 1), and competing heterogeneous firms with autonomous price setting capacities together constantly change production structures and make the population of firms endogenous.³ Therefore, Gerard Ballot and Erol Taymaz could demonstrate how new growth theory type“network externalities” may be created in MOSES markets by firms raiding each other for workers’ skills (see Section 4.4.4).

Since the MOSES model by prior design is selection based, initial state dependent and therefore highly nonlinear it is prevented from ever converging ex post onto a determinable exogenous equilibrium (“steady state”) path, or for that matter onto an analytically determinate stochastic equilibrium path.⁴ The MOSES model thus distinguishes itself positively from neoclassical orthodoxy that lacks markets, including, surprisingly, also from agent based models where explicit markets naturally belong, as noted by Ballot et al. (2015). With this empirically relevant specification, however, comes problems of theoretical intractability and parameter estimation.

Experimenting with the economy wide market dynamics of the MOSES model, for instance, has gradually told me to be skeptical about the (empirical) relevance of the ergodic or (exogenous) equilibrium doctrine underlying received economic modelling, and argued by Samuelson (1968) to characterize a good economic model. It has rather convinced me of the necessity to embrace micro simulation mathematics to break out of the confines of economic orthodoxy to understand how markets work in self-coordinating an expanding economy, and what kind of transactions costs are incurred. By embracing the microsimulation method I can thus see a future of enormous untapped opportunities to advance general economic theorizing, and without breaking with the spirit of classical economics, but only escape the inhibiting toolbox brought into economics with modern neo -Walrasian mathematics.

Just in passing, the “disequilibrium” character and initial state dependency of the MOSES model economy means that it must be empirically characterized before it can be subjected to general theoretical analysis. I see that as a relevant and therefore positive characteristic of an economic model. In the MOSES world there is no such thing as pure general economic theory.

In the MOSES evolutionary model approximation of what I call an Experimentally Organized Economy (EOE) price expectations govern economy wide behavior. Market price expectations are integrated within the individual business plans of profit seeking production agents (firms) that grow by bringing in global best practice technologies through their endogenous investmets. When attempting to realize these plans, agents must compete for resources in labor and financial markets and for income in product markets. Competition forces all business agents to innovate preemptively to stay in business and to engage in individual ex ante long term expected wealth maximizing dynamic games, that must be constantly revised. The outcome is a complex, constantly ongoing flux of winners and losers, economy wide success being associated with a dominance of the former, the latter being regarded as a normal transactions cost for economic development (Eliasson and Eliasson, 2005). Together it all amounts to a gigantic ex ante long run economy wide market self- coordination game propelled forward in time by production agents climbing ex ante profit hills, or if you like, as the entrepreneurial competition driven Schumpeterian Creative Destruction process stylized in Table 1 and embodied as the driving force in the MOSES model (Eliasson, 1995; Eliasson, 1996). Economic growth occurs if winners dominate over losers in this market game. Since the ex ante profit hills, as perceived by firms, change from period to period, because of all the competitive profit hill climbing going on, these peaks will never be attained. Through innovation nudges and competition markets both govern the allocation of resources, force production agents to innovate, and self-coordinate an economy wide evolution, which rarely comes even close to what may be maximum possible at each point in time. Forcing the model economy closer to such a state by speeding up market competition (through policy) will eventually create increasingly unpredictable markets, raise business planning mistakes, and cause unstable economic systems behavior and perhaps even collapse (Eliasson, 1991a). Stable long run macro growth still, however, requires a significant turnover of firms, a steady churning of microeconomic structures and reallocations of resources over markets. As the publications of this anthology demonstrate, a MOSES type model economy exhibits such dynamic complexity that it is a vain hope for a central coordinator, like the Walrasian auctioneer, to be sufficiently informed to do a better job than the markets (Eliasson, 1991a; Eliasson and Taymaz, 1992).

The MOSES model thus departs from both modern Neo Walrasian welfare theorizing and macro demand economics, and for that matter also from neo Schumpeterian economics, all being modelling without markets, and rather sides with Smith (1776) original treatise of the role of the invisible hand of markets in both self-coordinating the economy, and creating economic wealth and welfare. Dynamic markets must therefore be allowed back into economic systems understanding. The moral is that economics without markets is a bad idea (Eliasson, 2023a). This anthology should clarify technically what that perhaps surprising statement means.

My argument for the necessity of non-equilibrium modelling is not the same as rejecting the idea that markets are capable of successful economy wide self- equilibration. I only note, what I have repeated for half a century, that a different and more complex equilibrium concept is needed than the one we read about in economic text books (Eliasson, 1976a). Economy wide market self- equilibration or self-organization in the MOSES model only requires that economic evolution stays within a range bounded from above and does not collapse. It becomes socially and politically desirable that this bounded range is narrow and expanding. What the above summary of simulation experiments demonstrate is that long term evolution of the macro economy will all the time take place well below what is at each point in time physically feasible, and that a bounded range equal to zero (a “steady stationary state”) doesn’t exist as an operating domain of a realistically modelled economy.

As pointed out by Day (1986), initial state dependent nonlinear models will now and then enter phases of deterministic chaos. Such local economic environments, that the MOSES model occasionally generates, are at odds with social welfare ambitions of Western democratic governments. in fact observed that such economic models may have a determinate exogenous equilibrium, or an attractor, outside the operational domain of the economy, that it never comes even close to, and that the model economy does not have to converge towards this equilibrium.⁵ In MOSES terminology the ex ante optima mentioned above that each firm aims for, and revises from period to period will together constitute such a moving external attractor or “ex ante equilibrium” that constantly evades being reached ex post. It is therefore a mistake to aim policy at reaching it. These cosmological analogies may be a bit too abstract for empirically minded micro simulators, even though they might set creative imagination in motion. The same abstractions, however, relate to down to earth practical reality if the real world is nonlinear, selection based and initial state (history) dependent, which is more than likely. In such a world radically disturbed market environments occur now and then and make neoclassical welfare criteria based on achieving equilibrium states nonsensical. In the real nonlinear world they instead demand of policy makers that they not only understand what kind of disturbance the economy is heading for (a forecast), but also that they control both the short term and the long term outcomes of their interventions, which experience from MOSES modelling suggest that they will be uncapable of. The welfare interventions that such “experience” suggests, therefore, is rather to help individuals build social capital to cope with market disorder on their own (see Eliasson, 2009a and Eliasson, 2001 on social capital).

Another interesting example of MOSES applications needing abstract thinking is the use of modern financial tools, for instance the CAPM method developed by Sharpe (1964) to predict market volatility. The CAPM method was derived from a static Walras, Arrow & Debreu (WAD) type of general equilibrium model that has nothing to tell on a world out of equilibrium. Since the MOSES model is constantly “out of WAD equilibrium”, it became interesting to ask how reliable the Capital Market Pricing Model (CAPM) and its β coefficient were as a predictor of market risks when tested in the MOSES financial market environment and therefore by implication, in real world situations. This became possible when the new financial module of Eliasson and Taymaz, 2001 had been installed (see further Section 4.6 and Broström, 2003).

I have been constantly asked, from seminar to seminar, about the equilibrium properties of the MOSES model, and to say something about the seemingly subtle and unfamiliar policy problems associated with the unpredictable dimensions of MOSES economics. Static market clearing equilibrium where no agent has an incentive to move elsewhere doesn’t exist as an operating domain in MOSES. Mathematically, pushing it there will create an infinite regress and economic systems collapse. It is technically possible to speed up market processes such that financial markets come close to clearing in the sense that all firms that survive this competition earn the same rate of return, which is at the same time equal to the interest rate, a classical neoclassical equilibrium condition. We have done that and found that when all firms become almost equal (“representative”), diversity of structures is gone, markets have just about stopped functioning, and the MOSES model economy has begun to swing wildly before it collapses. So, these abstractions are not entirely uninteresting. Remember, that when the neoclassical model is on a steady state the rate of growth equals the interest rate. But by then the MOSES model economy has already collapsed (Eliasson, 1984; Eliasson, 1991b; Eliasson, 1991a; Eliasson, 1991c; Eliasson, 2005; Eliasson and Eliasson, 2005).

Welfare policy should thus focus on how markets function, not on achieving abstract market outcomes that cannot even be demonstrated to exist in a dynamic agent-based model with market self-coordination incurring unpredictable market transactions costs. It should come as no surprise that the integration of partial models by explicit market processes that make up the MOSES model economy has led to very different macro-outcomes than what each of the ingredient partial models would tell, as suggested already in Eliasson, 1976a). Macro models, for instance, to exist as economically meaningful artifacts, require such extreme restrictions of market functions, that micro to macro theorizing will take you down a dead alley if the ambition is to establish a market foundation for existing macro theory, or make it compatible with existing equilibrium modelling, a point made already by Weintraub (1979), and reluctantly admitted by Fisher (1972) and Fisher (1983).

To understand economy wide market self-coordination without the help of the Walrasian auctioneer or a central planner was an ambition of the MOSES project from the start. Centrally planned and free market scenarios can be enacted and compared within the MOSES model (see for instance Antonov and Trofimov, 1993).

An economy is populated by millions of heterogeneous agents with their own incompatible agendas that they express by having their plans tested in markets. Fundamental economics therefore is about modeling how markets sort out compromises from those generally incompatible plans, and what transactions cost are involved, not least in the form of economic mistakes being committed. It also places limits on central policy ambitions to govern the economy against the preferences of market agents, and to understand why pushing the wrong polices raise those transactions costs (Eliasson and Taymaz, 1992). To model the economy as a controllable economic system is therefore fundamentally wrong. Ballot et al. (2015) noted the absence of markets capable of economy wide self-coordination even in agent-based models and that attempts to model macroeconomic evolution without considering how agents interact in markets are a mistake.

While economists at Western universities seem to have had difficulties understanding that modelling economies as controllable systems without disturbing markets is fundamentally wrong, Soviet economists had first-hand experience of what it meant trying seriously. When the CEMI research institute in Moscow asked permission in the 1980s to use a stylized version of MOSES loaded with Soviet data to study the problems associated with transferring the centrally planned Soviet economy into a market economy (see below Section 5) they had understood that the MOSES model economy was market self-coordinated and featured a centrally planned economy as a special case. This illustrates how fundamental questions of economic theory and practical economics are closely integrated and can be studied using micro simulation methods. It also highlights the importance of empirically relevant model design such that the categories of the internal artificial reality of the model economy become empirically familiar and can be related to interesting aggregate targets. A richly specified and empirically grounded internal artificial reality gives MOSES the capacity to tell about possible, not yet observed or thought of, empirical properties of a complex industrial economy, and to formulate new hypotheses to test on empirical data (Eliasson, 2018). Regard the organization of this summary article as an attempt to clarify that philosophy.

Financial markets trade in information and are therefore central to the efficiency of the allocation of real resources, the self-coordination of the economy and hence also for achieving long-term stable economic growth. The introduction of sophisticated financial market arbitrage in MOSES, substituting a capital market regime with trading in financial derivatives for the previous industrial banking system (Rybczynski, 1993) came with Eliasson and Taymaz, 2001, and later elements of competence blocs with Ballot et al. (2006). This represented not only the first major upgrading of the MOSES model, but also illustrated the flexibility in both generalizing and calibrating the model that came with its modular design. Since price expectations are integrated within the business plans of firms this integration meant major change in the dynamic core of the model. Financial market transactions no longer took place independently of product and labor market transactions.

When firms subject their business plans to be tested against the incompatible plans of other firms, the markets sort out compromises and firms learn by feeding back their experience from such constantly ongoing market confrontations, revise their plans and try again (a Stockholm School feature). In the process business mistakes are made, prices and quantities are determined within and between periods (Eliasson, 1976b). The original modular design of MOSES was therefore instrumental in allowing the deep redesign of its core market mechanisms that the new capital market based financial system required. For instance, introducing financial derivatives that speed up market arbitrage , makes the MOSES economy more statically efficient, but may also disturb market arbitrage, reduce price predictability , and increase the rate of production and investment mistakes. Even though static efficiency is increased, long term macroeconomic growth may suffer, as simulation experiments show. Allowing insiders to inform markets by their trades, on the other hand, does increase long term growth (Eliasson and Taymaz, 2001. Up to a limit the innovation input of increased turnover of firms therefore raises long term sustainable growth. Business mistakes, however, eventually, as an increasing rate of structural change lowers market price predictability, begin to cumulate in numbers and sizes and to dominate the positive innovation push effects, and thus bend down the long term growth curve (Eliasson et al., 2005).

Competence bloc theory as formulated in Eliasson and Eliasson (1996) and Eliasson (2009a) introduced two types of economic mistakes in the allocation of investments; Type I mistakes occur when firms are late in identifying that an investment mistake has been committed, and in correcting it. Type II mistakes are incurred when firms fail to commercialize promising technical innovations. Type II mistakes can only be defined in MOSES type economy wide agent-based models in which markets, governed by human capital (“competence”), self-coordinate activities and allocate resources. During such much less than fully informed market selection winners may get lost for such long times that the loss becomes in practice permanent for the local economy. The more complete and diversified local competence blocs the lower the risk that winners be lost. The ultimate economy wide outcomes therefore depend on how market agents that perform that allocation are spontaneously organized to learn from experience to identify promising technologies (“winners”) and support their commercialization. Competence bloc theory explains that. Hence, business mistakes committed by market agents will embody information that can be learned to improve next period anticipations and plans. All that is represented in the MOSES model. Under such circumstances economic mistakes and their consequences should, of course, not be represented by stochastic processes, and the economy should not be modelled as a stationary process, a conclusion that brings up not only the absurdity of the rational expectations (RE) proposition, but also the relevance of the old distinction of Knight (1921) between risk and uncertainty (Eliasson, 2015). Competence bloc theory allows for winners to be permanently lost to the local economy, a possibility neoclassical economists tend to object to by arguing that in the end somebody else will come up with the same winning technology; an "ergodic" assumption of sorts.Perhaps, but it may be somewhere else and many years from now. The short period long term dynamics of the MOSES model in fact also embodies an answer to this. While winners may get lost (Type II mistakes), endogenous innovation,entry and commercialization in MOSES migh quite possibly generate long term winners that compensate for previously lost winners at some higherm level of aggregation. The point made in MOSES modelling is that the net flows of captured and lost winners over the longer term may be both positive and negative, and the difference will cumulate over time into sizable differences in economic wealth created, for instance, for a regional or national economy.

No serious study on tax and transfer heavy economies like the Swedish one can abstract from the public sector and its financing. The public sector in MOSES figures as a huge non- market economy superimposed on, and integrated within its core market economy through the large tax wedges associated with its financing, that bias price signaling in markets. The mere presence of this rigid lump in the midst of the market machinery thus raises the risk of distorted prices and concerns about market incentives and consequently of business mistakes. Even in this crude form as a producer of public services distributed free of charge, and a reallocator of incomes, the MOSES public sector is thus of great principal interest.⁶. The modular design of MOSES makes the integration in Eliasson (1980b), republished in this anthology, of taxes and transfers, and the placing of an inflexible non market public sector in the midst of MOSES markets, technically easy and the consequent accommodation problems empirically obvious (see Section 4.1).

Long before the concept of Type I and II mistakes had been formulated, Eliasson and Lindberg (1981) had demonstrated in MOSES experiments that even major investment mistakes (in their case caused by large tax wedges in the price system) had small economy wide effects if they were identified early as mistakes, and promptly shut down. The Swedish industrial support program of the 1970s and 1980s, when subjected to an economy wide cost benefit study on MOSES (see Section 4), became an extreme version of late understanding and failure to correct. The large economy wide costs were in both cases incurred by keeping workers locked up in mistaken, low productivity production, depriving the economy of their higher productivity inputs on other jobs.

Before moving on to the empirical validation of the MOSES model the theoretical bag must be tied up. MOSES has had an uneasy relationship with conventional Keynesian macro and neoclassical economics, and for both good and bad reasons. The latter both hinge on the equilibrium concepts of neoclassical orthodoxy, and the impossibility of making macroeconomics compatible with an economy self- coordinated by market agents with price setting autonomy. What makes MOSES economics depart from economic orthodoxy is the lack of markets in both. Making aggregation endogenous over dynamic markets, and breaking up static equilibrium to introduce Stockholm School ex ante -ex post economics at agent levels, not only provides an answer to the perennial question asked for instance by Arrow (1959) and Lindbeck (1966), who sets the prices when Walras auctioneer isn’t there to do it? Breaking up static equilibrium also puts life into economic theory.

The early versions of the MOSES model consisted of the micro based dynamic market core documented in Eliasson (1976b), that integrated Schumpeterian entrepreneurship with Stockholm School expectational feedbacks within a Keynesian demand structure such that all incomes generated, after having been taxed down, were run through a Stone type expenditure system estimated by Klevmarken and Dahlman (1971). The latter had been inserted in MOSES after some nonlinear features had been added (investment demand and demand for intermediate goods were modelled at firm level). Very soon thereafter the production structure of the model was expanded to be based on a complete eleven sector input output model restructured onto the OECD End Use classification to be compatible with the market categorization firms used in the micro data we collected in a separate survey, the Planning Survey of the Federation of Swedish Industries. Four sectors or subindustries could now be populated with (data on) real individual firms (see Appendix). From above, the model now came out as a familiar eleven sector Leontief – Keynes model (see Figure 2.1 in Eliasson, 1980a Republished in this Anthology).¹⁰ An estimated such sector model was at the time available from a separate project at the Industrial Institute for Economic and Social Research (IUI) that I had been involved in putting together and had access to. Several production sub industries (in MOSES currently five) in that sector structure could now be carved out and replaced by any number of heterogeneous and boundedly rational or “ignorant” firms in the sense of Simon (1955).¹¹ The individual firm data needed to put the model on an empirical footing was at the time not available, so a new Planning Survey project (Virin, 1976; Albrecht, 1978), was started at the Federation. Questions were formulated based on information from Eliasson (1976a) on what kind of data firms themselves based their decisions on, and designed to serve the needs of the model.¹² To establish a micro to macro and stock flow consistent initial state database was therefore of great principal importance. Firms in the MOSES model base their decisions on data that (1) reflect the "disequilibrium" situation in the real economy, but (2) are also afflicted by statsistical errors, most important the errors that depend on inconsistencies between public Nataional Accounts (NA) and Planning Survey data, that were concentrated in the synthetic firms created from the differences between NA data and the aggregated Planning Survey data for each market to scale the model up to NA level. It was importanrt that these statistical errors be minimized, since in non linear initial state dependent models of the MOSES type they tend to cumulate (see Appendix (Section A.1.5), and Albrecht et al. (1992) on how that high quality data base was constructed for the 1982 initial state).

Business agents invest, and with each new endogenous investment vintage new “global best practice technologies” from an exogenous global pool of technologies are brought into the production system of the firm (for details see Eliasson, 1976b). Supplies of individual firms are thus restricted from above by the labor and capital productivities of these best practice investment vintages that firms access through their endogenous investment decisions. With Carlsson et al. (1997); Ballot and Taymaz (1998); Ballot and Taymaz (1999), firms investing boldly in R&D could locally improve upon those globally available best practice technologies (innovations), improvements that were in turn made locally available through an upgrading of the global pool of best practice technologies in Ballot and Taymaz (2012).

The MOSES model was used early to clarify the relative importance of new technologies and markets in the commercializing and scale up of new technologies for a joint project between IUI and the Swedish Academy of Engineering Sciences (IVA). This so called “Grand Project” came in handily (for us) to obtain data on best practice technologies. In a survey to IVA’s members (mostly technology directors at Swedish firms) historic data on best practice technologies were collected that could be used to calibrate the MOSES model (see Carlsson et al., 1981 and IVA project in Section 4.2).

Business agents in MOSES with different production structures and market experiences make up their plans and decisions based on Stockholm School adaptive price anticipations with learning feedback. Expectations are always more or less mistaken (even “in expectation”). Ex ante - ex post differences therefore include economically valuable information that firms interpret (learning) and recycle as experience to improve next period (quarter) anticipations.¹³ This Stockholm School feature is a novel element of MOSES economics, that makes mistaken decisions the main market transactions cost incurred.¹⁴ On top of this feedback dynamic, markets in the MOSES model are constantly disturbed by Schumpeterian entrepreneurs that enter unexpectedly, causing Schumpeterian Creative Destruction on existing production structures and forcing failing incumbent actors to exit, leaving in their wake an endogenous population of business firms (Table 1). Expectations and economic incentives make firms invest and grow, and exert, together with Schumpeterian entry, a collective competitive pressure on all incumbents. Such competitive challenges force them to innovate preemptively (Eliasson, 1995Eliasson, 1996) through engaging in higher risk R&D investments (Carlsson et al., 1997; Ballot and Taymaz, 1998), to forestall being competed out of the market (exit). This ex ante - ex post dynamics is necessary and sufficient to keep market competition driving economic evolution forever, but not sufficient for economy wide systems stability.

Entrepreneurial entry, however, performs the dual role of both forcing incumbents to produce more efficiently (competition) and to upgrade production structures (innovation), thus maintaining structural diversity in MOSES simulations. The latter diversity enhancing function is the final dynamic element required to keep economic systems evolution resilient and not easily destabilized Eliasson, 1984; Eliasson, 1984; Eliasson, 1991b; Eliasson, 1991a; Eliasson, 1995). A complicating factor is that structural change may become too fast, reducing the reliability of information signaling in markets, and hence possibly growth by raising the incidence of business mistakes excessively. Cumulating transactions costs through escalating investment mistakes then eventually destabilize the economy and lower growth (Eliasson et al., 2005).

MOSES firms export a share of their products that depends on the relationship between the (exogenous) global price and the (endogenous) domestic price firms fetch (Eliasson, 1976b). In foreign markets MOSES firms are currently price takers.

Inputs in each market are similarly governed by relative price dependent macro import functions. Even though historic ex post simulations and parameter calibration are based on registered price data, general MOSES experiments must carefully define the relationships between foreign and domestic prices in each market such that they become compatible in the long run with global best practice productivity brought into firms through their endogenous investments. One could perhaps say that the global economy is assumed to be in approximate global steady state equilibrium. Exogenous foreign prices, including the global interest rate, therefore had to be exogenously calibrated in experiments such that firms investing in those best practice technologies would earn a return (on the margin) approximately equal to the global interest rate (Eliasson, 1983). The technically interested reader will observe that this exogenous specification is a “soft ergodic” assumption that has been sneaked into the model through the back door. I consider this a temporary deficiency that should be possible to remedy by calling in Ballot and Taymaz (2012), who have modeled a global economy of MOSES clones that are integrated by the diffusion of new technologies between the cloned economies. Firms in those cloned MOSES economies imitate superior firms in other cloned economies, a specification that can be more generally interpreted and modeled to include different kinds of interaction, including direct trade in technologies, foreign direct investments (FDI) and the movement of human capital by allowing firms to recruit across borders very much as already done within the MOSES model (see Eliasson, 1976b and sections 4.4.4 and 6.2 below). Since local innovations in individual firms will also upgrade the global technology pool, innovations in one economy will therefore globally diffuse through trade in technologies and be made available to firms in all (cloned) MOSES economies through their endogenous investments.

In studying the consequences for the MOSES model economy of the oil crises of the 1970s, oil price shocks were entered exogenously to disturb global market order. They thus hit the Swedish economy as unexpected price shocks in export and import markets that MOSES firms had to cope with, destabilizing the Swedish economy , and very much so, as the inflationary consequences diffused for years through the markets of the model economy (Eliasson, 1978a; Eliasson, 1978b).

The endogenization of local productivity development by Ballot and Taymaz (1998) opens interesting possibilities to escape the long run global market equilibrium or soft ergodicity assumption of the current MOSES design, and thus to endogenize global price determination. With productivities endogenized as in each local economy competition in local (cloned) markets and globally through international trade as before, global prices in each submarket will be endogenously determined. It will have to be a principally simple but dynamic design, that would still be preferable to the current static global “steady state assumptions”. This is an interesting proposition for the future (see Section 6.2).

3.4.1 Accounting stock flow consistency (SFC)

Fundamental to MOSES economics, or the Experimentally Organized Economy (EOE), is that agents’ price expectations and plans are never mutually consistent ex ante, but that markets are constantly at work trying to eliminate inconsistencies, however never arriving at a consistent equilibrium because the circumstances at each moment of this market arbitrage change the circumstances for the following moment. Trade is constantly taking place, and as agents register the transacted prices and quantities, they revise their expectations and plans and adjust inventories accordingly. Ex post registered accounts will therefore be framed ex post (however not ex ante) within a micro to macro, stock flow consistent (SFC) National Accounting (NA) framework Eliasson, 1976b. The MOSES model is SFC because it keeps track of all items in the balance sheets of all agents (firms, households, the bank, the government and macro sectors like construction) throughout a simulation. Nothing disappears from, or appears in the model. Albrecht (1992) documents the exacting work associated with compiling the micro macro, stock flow consietent initial state for MOSES for the year 1982. Indirectly this consistency is of courase reflected both in the manual of Eliasson (1976b) and in the original preudocode documented in Lindstenz (2023).

Figure 1 illustrates how each business system (firm) relates to the external market world. Shaded blocks tell where Stockholm School expectations (Wicksell), Keynesian demand, and Schumpeterian entrepreneurs enter. Table 1 explains the principles of entrepreneurial entry induced Schumpeterian Creative Destruction, and the core dynamics of the MOSES model economy. The complexity of the internal dynamics of one business agent depicted in Figure 1 is sufficient to tell that the agent is neither in full control of its internal economy, nor in control of its market environment composed of all other agents and Government. Simon’s bounded rationality prevails universally and forever.

Figure 1

Download asset Open asset

Besides being modelled bottom up on observed decision practices of business agents, and their known access to data (as discussed in the previous sections), a priority of the MOSES project has been that it be (1) a general representation of the market dynamics of an advanced industrial economy, and (2) empirical, and in that order. That ambition necessitated an agent and market selection based, initial state dependent and therefore highly nonlinear model composition. Initial state dependence meant that the model economy must be “empirically” positioned initially before theoretical analyses begin. For obvious reasons our choice was to implement the model on Swedish data.

Ponder the following model: Y(t)=F(X(t), Y(t-1)) + ε (t), where ε is a stochastic factor of sorts. Y(t) is a vector of dependent variables to be explained, X(t) the explanatory variables, and Y(t-1) a vector of lagged dependent variables. In nonlinear initial state dependent models {X(0), Y(t-1)} represents that initial state. The empirical evidence brought together in composing the MOSES model and accounted for in the previous text tell that nonlinear initial state (history) dependent models are in principle unavoidable in empirically relevant economic modelling of the longer term, and that the short term is very short in a model that runs by, and feeds back agents’ market experiences by quarter.

The standard assumption in econometric applications is (1) that we have got the specification F() right, such that the {ε} are independent of all X and lagged Y, ¹⁵ (2) that initial state measurements are of sufficient quality to warrant the assumption that any measurement errors are small and not systematically related to the {ε},¹⁶ and (3) that these model assumptions hold for both the estimation period and the future period for which the estimated model is applied, for instance to forecast. Micro simulation mathematics in principle allows modelling to come close to that ideal without compromising analytical/theoretical tractability. Another problem is that estimation difficulties increase with the relevance of the specification (see Sect.3.8 below).

It is important that {ε} includes no systematic component that explains some of Y, leaving something to learn from studying the {ε} (Samuelson, 1965). If it does the specification of F() should be expanded to account for that. The point made is that microsimulation is the perfect method to accommodate such expansions both for general theoretical analysis and empirical modelling. We nevertheless recognize that however complex the F() becomes, it will always be partial in some respects, and more or less violate the above assumptions. The stochastic characterization of estimated parameters becomes dependent on this misspecification. The MOSES model is however a perfect design to minimize this problem, in that it is quite general to begin with, but also that, for each experiment (and mainly because of its modular design), the “master model”, maintained as a “general platform” for experiments, can be customized to minimize the partiality problem in each application.

Alfred Marshall is credited with coming up with the idea of partial analysis and that the economist can say something empirically meaningful under the ceteris paribus clause of holding some endogenous circumstances constant. That helped usher in strict formal analysis of clearly circumscribed problems in economics. There are however limits to how far one can go by invoking the ceteris paribus clause without engaging in nonsense economics.

Partial models, used to interpret complex interdependent real problems, provide interesting illustrations. See for instance section 4.4.3. The MOSES model therefore is a serious attempt to minimize the partiality dilemma, and a calibrated MOSES model can in prinicple always be used as a "docking station" for partial models to test for the biases that come with artificially holding endogenous variables constant.

All micro based macro models I know about are static in that they neither allow for sequential (periodic) feed backs of ex ante - ex post differences, for instance business mistakes, nor allow agents to learn from the information that is embodied in their mistaken expectations to improve next period plans (learning). Such micro simulation models are neither autonomous in the sense of Haavelmo (1944), nor stable over time, and thus in principle dynamically mis -specified. Even though this deficiency may not matter for (very) short term “static” projections, MOSES simulations beyond a few quarters tell that the consequences of such misspecifications often begin to cumulate. Since the ambition of the MOSES project was to come up with an autonomous model that would be useful for historic counterfactual studies (“What would have happened if instead …Eliasson, 1976b) this problem, typical of selection based initial state dependent models, had to be squarely faced. But autonomy is a complicated requirement, and we soon discovered one problem that was common to all our long term simulation studies, namely that autonomy in a model economy self-coordinated by competing market agents required that structural diversity be endogenously maintained over time (Eliasson, 1984; Eliasson, 1991b; Eliasson, 1991a; Eliasson, 2023a). Without entrepreneurial entry of innovative firms markets eventually stopped functioning. Not only did competition cease, but structural diversity was reduced because of exit, and often to the extent that the entire model economy became increasingly unstable and collapse prone. This is probably the best example of how the MOSES model economy has been constantly empirically validated by identifying the origin of unfamiliar economy wide behavior (or surprise economics Eliasson, 2018), and if found empirically unreasonable, caused modification of model specifications.

By embracing the micro simulation method, we have made it possible not only to upgrade economic model specification) such that the independence of {X, Y(t-1)} and a stochastic {ε} will become increasingly more empirically reasonable. The standard prior assumption that F() is not dated to the historic data on which it has been estimated, but also valid for both counterfactual studies on that historic period, and predictions for another chosen, perhaps future period , can be made more plausible. In addition we have taken a serious step in integrating economic modeling and measurement. With that the empirical relevance of prior model specification (of F()) becomes of supreme importance, making all models (necessarily) problem dependent. With F() customized to address a defined decision problem, and the empirical credibility of F() conditioned by the empirical relevance of the priors that have gone into that customization of F(), the decision maker who uses the model, and who might have designed the model, therefore must also know its properties such that s/he is able to interpret simulation results in terms of the internal artificial reality of the model (See Section 5). For you, as a decision maker, the model design you use defines your understanding of the decision problem you are facing. This is the way Simon (1955) and Simon (1957) defined bounded rationality. Eliasson (1992) uses Simon’s concept to define the limited understanding of any decision maker, be it MOSES firm managers or policy makers using MOSES to understand the economy. This problem becomes more difficult if you want to convince an outsider of the empirical credibility of your understanding, and by implication of your simulation results (See further discussion in Section 5).

Assumptions (1), (2) and (3) therefore integrate the economic principles underlying the empirical model inthe sense that relevant circumstances of the decision to be taken have been accounted for to the extent that other factors (“safely”) can be assumed to affect F() as independent stochastic noise.¹⁷ The strategic underpinning of the MOSES project therefore is that micro to macro modelling embodies the potential to move specification of the model as close as possible to satisfying the above assumptions, and by far much closer than is the case with construed macro or constrained static equilibrium models, and that today mathematical simulation (a) makes general theoretical analysis of complex micro based macro models possible using numerical methods, and frees the analysts of the limitations of conventional mathematics.¹⁸ Of particular importance is the possibility of overcoming the partiality predicament (Sect.3.6), that will be further exemplified below (Sect.4.4.3). Computer based simulation mathematics thus overcomes many difficulties of complex formal analysis. Second, (b) the micro to macro or microsimulation method allows efficient use of available micro data, that is left unused in artificial statistical aggregates, a point made long ago by Orcutt (1957) and Orcutt (1960). Micro simulation modeling thus not only is a design to endogenize aggregation over explicitly, and more realistically modelled markets than the extreme assumptions about markets underlying macro models and equilibrium models.¹⁹ It also (c) offers a design potential to increasingly satisfy the stochastic assumptions (1), (2) and (3) above on the properties of {ε}, needed for the parameters of the model Y(t)=F(X(t), Y(t-1)) + ε (t) to be effectively estimated. Computer based micro simulation mathematics thus offers a method to integrate sophisticated general economic theorizing with economic measurement that has yet to be taken full advantage of.

There is a corollary to the last statement, namely that to satisfy the condition of independence of F() and factors not recognized, raises model complexity in practically all decision situations. While the microsimulation method makes general (numerically based) theoretical analysis possible, with functional complexity comes mounting difficulties of parameter estimation. The more empirically relevant the model design, the more likely the desired stochastic assumptions of {ε} will be satisfied, but the more difficult it will be to determine the stochastic properties of the estimated /calibrated parameters. Even though availability of data, improved measurement of initial conditions and computer capacity may eventually overcome this dilemma (Klevmarken, 1978; Klevmarken, Anders,2022, n.d.), this doesn’t solve the current problems of MOSES estimation, or for that matter of all agent based non ergodic models (Richiardi, 2017; Grazzini and Richiardi, 2014). Hence, to be up to its ambition to be empirically relevant, the MOSES project has had to do with calibration as a second-best approach to estimating its parameters.²⁰ Calibration, as I define it, differs from estimation in that the uniqueness and the stochastic properties of the calculated parameters have not been determined.

Ordinary least squares (OLS) applied to linear models give consistent estimates that are asymptotically normally distributed. MOSES is far from linear and its loss function therefore highly complex and not analytically determinate. It should however be possible in principle to figure out, using numerical methods, what kind of estimates that will be obtained when the sum of squared distances between simulated and control variables is minimized using the computer based calibration program of Taymaz (1991b). This loss function in MOSES is however very complex, and its selection features mean that minimization may result in several local minima. This problem, however, also afflicts standard estimation methods, and the only way to decide which parameter settings to choose is to do what we have already done, namely to validate the calibrated parameters against other exogenous facts.

On this Hansen and Heckman (1996) state comfortingly that “there is nothing sacred about the traditional loss functions in econometrics associated with standard methods, like OLS, except that they can be “rigorously justified”, producing, for instance, in linear models, consistent parameter estimates that are asymptotically normally distributed. So far, we are therefore sufficiently satisfied with the set of calibrated parameters using the method of Taymaz (1991b) and Taymaz (1993) to refer to results from MOSES simulations as relevant for the dynamics of a Sweden like industrial economy. The question is rather what steps need to be taken, if needed, from there, and before data and computer capacity have become available, to obtain a sufficiently rigorous representation of the Swedish economy? My argument on this has been that few economy wide models satisfy that requirement, being both mis specified, and failing to provide a convincing argument that their ε (t) have the required stochastic properties.

On this MOSES model validation offers the following advantages.

The specification possibilities offered by the microsimulation method have made it possible to substitute direct measurement for parameters in many places (initial state, bankruptcy law and exit), reducing the need to estimate for instance exit functions and distributed lags etc, thus also reducing the number of parameters to be estimated to the critical couple of dozens that regulate market self-coordination dynamics. On them we have found that they have stayed quite close from model upgrade to model upgrade and from changing from the manual calibration method of Eliasson (1978a); Eliasson (1978b); and Eliasson and Olavi (1978) to the Taymaz (1991b) and Taymaz (1993) computer calibration method. We take that as a comforting sign of consistent calibration (for details see Appendix).

The annual Planning Survey has so far produced panel time series data of individual firms that can be used each new year together with national accounts (NA) statistics, as fresh test data. The most obvious procedure would be to retest the model each year as new data becomes available. Under the assumptions (1), (2) and (3) above, and that the specification of the model holds for each future period added, one would then expect new calibrated parameters to cluster within a narrowing interval. If supported from year to year this test would be as good as any estimation procedure, and with increasing confidence warrant references to the Swedish economy.

Finally, model specification will never be general enough to hold forever, and calibrated parameters may eventually begin to stray off in new directions. This should prompt us to look for structural changes in the real economy that the model has not picked up, and to modify its specification. One example of this validation method is that when the model economy began to exhibit severe instabilities in long run simulations, and especially when some of the critical market parameters were changed from the calibrated values to speed up market arbitrage (Eliasson, 1984; Eliasson, 1991b; Eliasson, 1991a; Eliasson, 1991c). This is not a desired property of a good model. The reason, we soon discovered, was that with entry shut off structural diversity deteriorated rapidly when faster market arbitrage, notably in the labor market, forced inferior firms to exit faster and faster. Structural diversity and systems stability were restored when the Hanson (1989); Taymaz (1991a) and Ballot et al. (2006) endogenous entry algorithms had been introduced in the MOSES economy (Eliasson, 1991a). In Section 6, I discuss extensions of MOSES specifications , and why, because of its modular design, some of these design changes may not much affect the parameters of the core market innovation and competition machinery.

Empirically well-grounded a priori specification with the ambition to satisfy assumptions (1), (2) and (3) thus should always come before estimation. It is therefore bad scientific procedure to throw in a standard model without comment except perhaps for a reference to its general theoretical familiarity. The basic assumption of MOSES calibration has been that we have got the specification right, such that the simulated results can be confidently interpreted in terms of the internal artificial reality of the model economy to support the decision for which it has been customized. If the specification is good and parameters reasonable you should feel confident in the simulated results. Micro to macro or micro simulation modelling thus offers the advantage of both allowing and forcing the economist to come up with empirically well-grounded prior specifications that are as close to satisfying assumptions (1), (2) and (3) as is possible. (Since the modular design of the MOSES model technically simplifies a customized design of the model to be consonant with the decision problem addressed, a carefully calibrated MOSES model in no way is less empirically credible than a perfectly estimated but falsely specified model, for which a careful analysis of the biases of an unsatisfied assumption (3) has not been documented. Both rest on a priori assumptions that must be compared and argued separately, and before estimation, and therefore suffer from the same problem. The wrongly specified partial models discussed in Section 4.4.3 are examples).

We argue (1) that MOSES specification is as empirically relevant as has been possible to achieve for an advanced industrial economy, and that its original modular design has made gradual expansions possible. Second (2), the specification of the MOSES model, trimmed of detail down to its core micro-market- macro dynamics, as specified already in Eliasson (1976b), has allowed the general theoretical analyses of its dynamic properties commented on in the first two sections of this document. Third (3), in setting MOSES up for simulation studies it has been loaded with initial state data of production agents (firms and divisions in large firms) from (3a) a separate so-called Planning Survey conducted annually by the Federation of Swedish Industries. The firm data (3b) in turn has been integrated stock flow and micro to macro consistently with the official National Accounts (NA) for Sweden. The latter serves as a size reference for the model economy. How this MOSES Data Base has been composed for 1982 is described in detail in Albrecht (1992). (1) and (3) are sufficient to refer to MOSES as representing a Sweden like industrial economy. The MOSES model, however, also claims to have the potential to represent the sequential dynamics of the real Swedish economy over the longer term. That, however, also requires (4) that a couple of dozen critical parameters that regulate the intensity of market competition, and the period to period Stockholm School dynamics of the model economy be properly estimated.

Already in Eliasson and Olavi (1978), about 20 so called “time reaction parameters” were identified that regulated either the longer term growth trends of the model economy, or the cyclical variations around those trends, or both. Since micro distributional data were collected in the Planning Survey it became gradually possible also to test the model’s capacity to generate historic distributional patterns (see Taymaz, 1992a and Appendix).

Over the years a three-step calibration/estimation procedure for these parameters has been developed:

1. Calibrate a selection of parameters against a selection of test variables, using the distance (sum of squared deviations) minimizing methods of Eliasson (1978a); Eliasson (1978b); Eliasson and Olavi (1978); Taymaz (1991a) and Taymaz, (1993), holding the exogenous variables at their observed values (notably foreign prices, interest rates, and productivities of best practice globally available technologies. On the latter see the IVA project in Section 4.2).

2. If several different calibrated parameter combinations with roughly the same fit are obtained, exclude obviously false ones by prior judgement, and regard the others provisionally as empirically acceptable, even if lying within a wide “confidence interval”.

3. Keep recalibrating the model as under A and B as new test data arrives and take it as a sign of “consistent calibration” if the calibrated parameters do not change much or converge into a multidimensional cluster of a priori reasonable values.²¹

These are principal criteria that have been intuitively followed over the years, but that remain to be worked out more precisely mathematically, something that makes sense, since the stochastic assumptions (1), (2) and (3) above are the same as those of standard estimation methods.

Our experience has been that a dozen calibrated parameters that regulate the core market dynamics of the entire model economy, represented by about the same number of macro test variables, as shown in Eliasson, 1976a) and in the Appendix, have stayed reasonably constant between both model generations and data base changes. The early manual method described in Eliasson (1978a); Eliasson (1978b); Eliasson and Olavi (1978) did not allow a wide search for good fits, so we could not be sure that the empirically correct combination of parameters did not lie elsewhere. With Taymaz (1991a) and Taymaz, 1993 computerized calibration method a much wider search became possible, using new test data. We were then happy to find that the new core model parameter estimates stayed roughly the same as the old estimates, even though the new estimates were obtained for a much-expanded model. Fortunately, we did not find any outliers with good fits.²² With a sufficiently long panel data base from the Panning survey an even larger number of parameters should make it possible to fit MOSES simulations to test variables representing distributional characteristics of firms.²³

Empirical studies on the MOSES model require that a full scale data base for the initial year be set up. This is a major statistical job, that was needed for the cost benefit study of the industrial support program of the 1970s, which also had a generous budget (see Sect. 4.3). We have, however, developed a method to keep the model constantly ready for empirically less ambitious studies. From a carefully prepared initial state, for instance the one for for 1982 (see Albrecht, 1992), the model can be simulated forward “guided by” known exogenous variables. It can then be reinitiated by substituting the most recent firm data from the annual Planning Survey for the simulated firm data, within the simulated macro frame of the model. This procedure makes it possible to keep the MOSES model constantly ready for “semi-empirical” simulation studies that require an updated initial state data base. For principal analyses of this kind it is not necessary to have access to new Planning Survey firm data. Model simulations generate complete micro to macro and stock flow consistent initial states for each future quarter. For theoretical studies these artificial initial states are in fact to be preferred to a complete reinstallation of MOSES on a new data base, since then all errors due to inconsistencies between Planning Survey firm data and official National Accounts will have to be fixed again. The simulated initial states can also be customized for the study to be done by varying exogenous variables or parameters etc such that a desired initial state is obtained, for instance a balanced macro-economic situation to eliminate the effects of cyclical factors over the longer run.

The MOSES model generates complete initial states for every quarter during a simulation. Technically it can therefore generate data of immense detail ”on demand”. It has however not been designed for predictions of such detail. A detailed short term “static” simulation, one or perhaps even two years ahead, might make sense, because then the sequential feedback dynamics mentioned above has not yet much affected the outcome.²⁴ Most empirical microsimulation studies I know about are “static” in that sense, and aim for distributional predictions of great detail, for instance the distributional effects on disposable income of a modification of the income tax and rent subsidy system, holding incomes unchanged. This would however be a misspecification for anything beyond the very short run, and MOSES experience from dynamic simulations does not recommend such static calculation methods. Since the MOSES model has only been calibrated on a selection of critical target variables that are directly affected by the feed back dynamics just mentioned, but routinely “predicts” many more, the principal problem of how to interpret the latter remains. Since these secondary dependent variables are tightly (linearly) related to the critical variables against which the model has been calibrated, difficult to understand extreme behavior of these variables will therefore, as a rule, be associated with extreme behavior of the test variables. Some such extreme output from model simulations tabulated in the policy study Eliasson (1992); Eliasson (1992b); Eliasson and Taymaz (1992) was criticized. On close inspection, however, the extreme model behavior brought up for discussion had been generated by the extreme policies the model economy had been subjected to (see further Postscript appended to the republished study in this Anthology).

Testing long term economy wide behavior of the MOSES model economy for sensitivity to external conditions, initial state measurement errors or parameter changes, using Monte Carlo experiments, is the final validation stage, and became routine practice in pace with the rapidly increasing computer capacity (see for instance Eliasson and Taymaz, 2000). Such tests frequently led to changes in model specification, and made us look for inconsistencies in data inputs, and suggested modification of some calibrated parameters. An absolute requirement set up for the MOSES model has been that the economy wide behavior of the calibrated version stays viable over the very long run. Collapse like behavior of the calibrated version over the very long run has always been a cause for concern and made us identify the reason. The most important example of this theoretical or empirical validation (See Section 2.3) was that the MOSES economy tended to exhibit strong fluctuations and collapse like behavior in some experiments. Sometimes it occurred when we allowed the model to run into the very long term in a short-term empirical experiment to see what happened. Sometimes we provoked the model by changing its parameter setting to force markets to reduce inefficiencies. The latter were theoretical experiments conducted in response to academic criticisms that we had not properly clarified the equilibrium properties of the model. We then found that a static market clearing equilibrium state of the model was unattainable, and for good theoretical reasons (Eliasson, 1991b; Eliasson, 1991b Sect.2.3 above), but also that collapse like behavior of the model economy should be considered normal under extreme conditions. The typical reason was a rapidly deteriorating diversity of structures and the convergence of a shrinking firm population towards similar (“representative”) firms, making markets dysfunctional. Turning on a closed down entry module and endogenizing entry remedied this model deficiency.²⁵ An important consequence of these theoretical experiments was that we started to study Planning Survey firm data more closely, and discovered that Swedish manufacturing had experienced a deteriorating structural diversity for many years due to reduced entry of new firms. We alerted Swedish policy makers and academics to the macroeconomic economic risks inherent in that development, but with mixed success (see Section 5.4).

Calibrated parameter estimates judged to be empirically acceptable, after having passed the above test criteria, in my view makes the MOSES model a dynamically and empirically acceptable representation of a Sweden like industrial economy. There is therefore all reason to take simulated outcomes seriously as occurrences that may take place in a Sweden like industrial economy under the circumstances defined by simulation experiments.

This section has been concerned with the validation of the MOSES model economy for empirical studies, in practice to make it represent a Sweden like advanced industrial economy, but ultimately , when proper parameter estimation has been achieved, to credibly represent the Swedish economy. Even the lower ambition level is sufficiently demanding on database quality to cause “maintenance” problems. Yet the MOSES model had survived for fifty years by 2024 and exists today in several operationally active versions that have been set up for particular applications (see Appendix). Let me therefore conclude this section by explaining how.

A master version of the MOSES model is kept installed on at least one database. It has been regularly calibrated. The model specification has now and then been updated and validated. The Appenix shows the history of MOSES progress in that respect. With time, however, and depending on what problems are addressed, a more radical updating of specifications will be needed for serious empirical studies (see section 6). Even though older versions of the model and old data bases retain sufficient structural and dynamic market characteristics to make them interesting for general studies of the properties of advanced industrial economies, with time the characterization “Sweden like” gradually loses its validity. Thus, for instance, the role of markets in long run macroeconomic evolution is still unique to the MOSES model economy. Simulation studies on a Sweden like industrial economy during the first half century after WWII will keep being historically interesting, especially since it is only a budget question to update the database and recalibrate the model (see section 6). Let me therefore explain how MOSES has been kept operationally active for general theoretical, but also empirically interesting general analyses of agent based macroeconomic market dynamics.

No serious study on the Swedish economy should be conducted without due consideration of its enormous tax and transfer system, that lodges deeply and heavily in the self-organizing price mechanisms of the markets. While the first two sections of this Anthology have been devoted to clarifying the principal functioning of the self-organizing market dynamics of the model economy, empirical applications were the ultimate ambition. For that reason, the complete Swedish tax and transfer system was introduced in the MOSES economy already with Eliasson (1980a). It was there demonstrated how tax wedges in the price system bias long term allocational outcomes, why the same tax take to finance the public sector and transfer payments using a payroll or a value added tax had different distributional and macro-economic consequences, and how a transfer from the one to the other incurred social costs on the economy in the form of a permanent loss of output. Simulating the macro-economic effects of corporate taxes and fiscal depreciation rules Eliasson and Lindberg (1981) demonstrated why major investment mistakes caused by tax wedges, or misdirected subsidies caused little harm to the macro economy if identified early and promptly closed. The large macroeconomic, “social” costs were incurred if production was carried on in the mistaken investments, a consequence that became all too familiar in the industrial support program of the 1970s and 1980s (see Section 4.3 below). Without the complete tax and income transfer module introduced with Eliasson (1980a) the next few empirical studies would not have been possible.

In 1977 I became the director of the Industrial Institute for Economic and social Research (IUI) and took the MOSES model with me. The same year IUI teamed up with the Swedish Academy of Engineering Sciences (IVA) to clarify, on contract with the Swedish Technical Development Board (STU, a government authority subordinated the Ministry of Industry), the technical and industrial competence endowment of Swedish manufacturing, and by implication the role of new technologies in Swedish economic growth in the past, and for the future. The MOSES model came to be used as a cognitive instrument to sort out different explanations of how the post oil crises stagnation had been caused, and to suggest policy solutions to get the economy out of its distress. As part of this so called “Grand Project”, IVA’s members, most of them technical directors in large Swedish firms, could be tapped for quantitative information about the historic development of best practice technologies. In addition to being of great interest to the” Grand Project” when interpreted in terms of the model, this data was used both to run historic experiments, and to calibrate the MOSES model.³² This project provided research results for an international IUI conference 1977 published as Carlsson et al., (1978) and Carlsson and Olavi, (1978), and caught the attention of the Directorate of Science, Technology and Innovation of the OECD, which asked for complementary simulation studies on the role of human capital, and education in economic growth, and commercialization competence. Several publications followed (Carlsson et al., 1979; Carlsson, 1972; Carlsson et al., 1981; Eliasson, 1979; Eliasson, 1980b; Eliasson, 1981).³³ This time the main contribution of the MOSES model was improved “cognitive insight” rather than micro to macro quantification. The economy wide micro market format of the MOSES model made the integration of technology and economics through commercializing markets a central concern. It kept us from getting locked up in the partial tunnel vision of the linear technology growth drive proposition attributed to Schumpeter (1942) that had become popular policy among stagnating economies on which Keynesian demand had been found not to work. The story now became, subsidize R&D investment and growth will return. Of course this did not work, since it neglected the costly complementary commercialization phase of identifying and scaling up technologies, as well as the complementary human capital needed. We found (see Section 2.1 on competence bloc theory) our focus on a more market minded Schumpeter (1911) to be more promising. (From the perspective of economic doctrines it is however interesting to observe that the linear Schumpeter of 1942, that grew into the “national innovation systems“proposition of Freeman (1974); Freeman (1987); Lundvall (1992); Nelson (1992) and Nelson (1993),³⁴ also branched out into a neoclassical version called New Growth theory, first formulated by Romer (1986). This time R&D fueled macro innovation functions created network productivity externalities and then economic growth. From our point of view it is interesting to note that the network externalities are represented in new growth theory models, for instance Romer (1986) and Jones and Williams (1998) by an assumed positive macro relationship between a general collective knowledge capital and all other factors of production (an externality, see Eliasson, 1996). Those network externalities can be simulated on MOSES as an endogenized consequence of financial market resource reallocations (Eliasson, 1976b), and/or reallocations of the human capital of trained workers (see Section 4.4.4 below). Without complementary financial market and human capital support, we found, little growth should be expected from new technology creation. On the other hand, improved labor and financial market performance, holding technology constant (No change in best practice new technologies) was found to significantly raise macro productivity growth over the medium term, until the reallocation effects of improved market selection had petered out (Eliasson, 1979; Eliasson, 1981). Similarly, simulation experiments gave no support for worries about technological unemployment. Job losses in stagnant industries caused by competition from technologically superior firms, that in turn caused a general macroeconomic expansion were compensated for by new job creations, accompanied by a temporary rise in transactions unemployment, as workers moved to new jobs over the labor market. In fact, if markets did not do that job, and/or if defunct producers did not shut down (Section 4.3) the technology induced expansion might fail altogether to come about. A series of MOSES simulations demonstrated that growth and employment outcomes required significant complementary structural change (Carlsson et al., 1981). Sustainable long term growth was in general impossible to achieve without significant complementary churning of production structures, increases in investments in new and profitable firms, exits of inferior producers, and significant accompanying movements of workers over the market from low performing to fast growing firms. Again, the labor market had to function as a market for these positive results to materialize. (These conclusions may sound self-evident, and they are, but MOSES simulation results played a role in preventing simplistic (partial) stories from dominating the economic policy discussion in Sweden, and in particular the proposition, pushed by unions, that manufacturing firms with problems should be subsidized to train their workers to be capable of “reorganizing themselves from within” without forcing workers to change jobs over the market. The next MOSES project tells a sad story on that theme).

The potential of MOSES for economy wide dynamic model based cost benefit studies of major policy interventions in the economy was soon discovered. A contract was offered by the Department of Industry to investigate the potential of the Swedish shipyard and standard steel industries to recover as viable businesses from the deep crises brought on them by the oil price shocks of the 1970s, and devastating Japanese Government subsidizing of its shipyards. (This required both a preparatory analysis of the industry, and setting the model up for quantitative studies. The first part was already available with the IUI study Bentzel et al. (1970), contracted by the Department of Industry, in which I had participated with a financial analysis of the six shipyards ((Eliasson, 1970). We knew that the Swedish shipyards were technologically more advanced, but several of them (the six together making the Swedish shipyard industry the second largest in the world at the time, after Japan), had unfortunately invested themselves into the wrong niche of the oil tanker market. On that Bentzel et al. (1970) had concluded that the mistaken technology focus and Japanese competition were not sufficient reason to allow the technologically most advanced industry to go, even though my financial analysis had found the shipyards in a dire financial state because of the competition from the subsidized Japanese shipyards. They would immediately be bankrupted if government subsidies were not forthcoming).

Some years along the extremely costly subsidy path the Swedish Government had embarked upon to save the large Swedish shipyard industry, IUI was asked again to evaluate the program historically, and for the future if it was continued. We were specifically asked to use the MOSES model for quantification, even though it was not ready and empirically properly calibrated for such politically delicate economy wide cost benefit analyses. All Swedish shipyards were however in the MOSES firm database, and we got access to a wealth of firm data, not least the exact disbursement of subsidies by quarter on each shipyard. The study has been documented in detail in Carlsson et al. (1981) in Swedish). I devote extra space to this dynamic economy wide cost benefit study because it is principally interesting as an application of dynamic MOSES type microsimulation method, and because it has been subject to an ex post evaluation in Carlsson et al. (2018), that was possible only because all computer print outs had been carefully archived in the Swedish Centre for Industrial History.

(The subsidizing of their shipyards by Japan to achieve dominance in both ship production and global bulk shipping of oil was of course not acceptable, as are not similar attempts by China today. Neither was it very clever of Swedish Government to make Swedish taxpayers subsidize the highest paid workers in industry to produce in practice zero value added for almost a decade, depriving the rest of industry of their work input, and pushing up wages, generally slowing economic growth elsewhere). In Carlsson et al. (1981) and Carlsson (1983) different alternative ways of distributing the subsidy money (the public budget remaining unchanged) were simulated on MOSES and compared with what was actually decided in the Government budget to be done well into the 1990s under the official prediction of exogenous variables as published in the Government Long Term Survey (“Långtidsutredningen”). The optimal policy turned out to be to terminate the subsidy program immediately, and disburse the subsidy money back to all Swedish firms in the form of a flat rate lowering of the payroll tax.³⁵ When subsidies were discontinued the shipyards promptly went bankrupt (depleted their net worth and exited), and Swedish manufacturing returned to the previous faster growth trend that the subsidies had forced it off. In the interim period it had however dropped to a lower level of output that had not been recovered at the end of simulations (Carlsson et al., 2018). The main reason for that was that the subsidies had both locked skilled workers into zero value added production, depriving the rest of manufacturing of their work input, and raised the general wage level, thus discouraging firms in general from investing and expanding. Shipyard workers were forced into the market when subsidies were terminated in 1985, but in a couple of years they were all (in model simulations) reemployed elsewhere, albeit at lower wages. That the speed of market correction so estimated is not that unrealistic witness the ex post assessment of the study in Carlsson et al., 2018. When Government abandoned its previous employment protection policies, allowing the labor market in particular to conduct its allocation functions better from 1995, industrial structures promptly began to reorganize, and production growth shot up with statistically negligible adverse consequences for unemployment. The main cost to Swedish society was a lowering of the level of economic wellbeing (GNP/capita) that the economy had not been able to recoup by 1995, which was very much what model simulations had also demonstrated).

The publications summarized in Sections 1 and 2 above are studies on the internal artificial realities of the MOSES model economy, and what different market organizations generally mean for macro. When the MOSES model was used in the late 1970s to evaluate the economy wide costs and benefits of the ongoing industrial support program the Planning Survey based initial state had just been installed (for 1976) , but the couple of dozen parameters that regulated the long term sequential market feed backs on firms had at that time only been calibrated using the hands on manual procedure of Eliasson (1978a); Eliasson (1978b); Eliasson and Olavi (1978) (also see section 3.10). We were therefore reluctant to push the politically sensitive empirical results to hard, although in retrospect mistakenly so (see follow up study in Carlsson et al., 2018). Only with the Taymaz (1991b) calibration program, and new specifications of the model we more confidently began to refer to results from MOSES studies as being valid for a Sweden like industrial economy.

In a number of studies Gerard Ballot and Erol Taymaz have taken the original, calibrated model as given, added new modules, recalibrated the model and, using the new computer-based calibration program Taymaz (1991b), and set MOSES up for empirical human capital based inquiries. I therefore place them among the empirical studies of this section.

Technically, labor in MOSES is homogeneous, but depending on what machinery workers operate labour productivities vary widely across the firm population (see Salter curves in Figure 2 in Eliasson, 1991a and in Figure 1B in Eliasson and Taymaz, 2001). This coupling of workers’ productivities with the machine capital of the employer was an acceptable approximation when the MOSES project began, and before, but has become increasingly wrong with the advent of a New Economy with a large and growing share of human capital intensive service production that, depending on market circumstances, can be conducted within a hierarchy, or be out contracted in the market. In a series of publications Gerard Ballot and Erol Taymaz have improved upon MOSES specification of human capital input in productionin by introducing both general and specific training of workers, to make labor in practice heterogeneous. Even though individual workers could not yet be followed as they moved from firm to firm, the model kept track of the average human capital stock of the firm, and how it was affected by general and specific training of its workers, and how that stock changed when its trained workers moved to other firms (Ballot and Taymaz, 1993; Ballot and Taymaz, 1996).

The major channel for technology diffusion is the market for workers with superior human capital. By raiding competitors for new technologies embodied in human capital the raiding firm taps into the collective body of talent of that market. The more of that talent available the more individual (raiding) firms benefit, raising both the collective body of talent through market diffusion, and aggregate productivity. The more firms train their workers (even at the risk of being raided) the larger the collective talent capital available for raiding Ballot and Taymaz (2001) note. Raiding is a labor market function. Enhanced training spurred by fear of competition to raise rates of innovation and/or imitation, thus also raises aggregate productivity and makes firms on average more competitive. The more labor market competition in MOSES the larger the networking externality that Lucas (1988) and Romer (1986) introduced as (hypothesized) positive correlations between a collective measure of general human capital, and all factors of production (positive externalities) in their macro production functions (see Eliasson, 1996 and Section 4.2 above). Ballot and Taymaz (2000) however also explain how (!!!) that same externality may arise through enhanced labor market raiding of firms for talent that forces each firm to innovate by engaging in more general training, or risk being competed out of the market, and to innovate more to be able to counter raiding firms’ wage offers. This market-based explanation is principally different from how econometric results on new growth theory macro models (that regress total factor productivity growth on a measure of collective R&D capital) are interpreted. Since the interpretations are different, they also suggest different policies. (It is of additional interest to note that this explanation of the market origin of network “externalities” could in principle have been investigated already on the original 1976 version of the MOSES model, although at the time we did not understand that theoretical potential. The fear of being raided explanation of innovation and endogenous growth was expressed differently in Eliasson (1995) as precautionary innovation spurred by fear of being competed out of the market. Together these two mechanisms suggest a market explanation of the network externalities statistically measured in macro new growth theory models.

Network externalities explain firm formation in that hierarchical organization beats the market in organizing production effectively. Competence Bloc theory (in Sect. 2) defines a cluster of financial market intermediaries that together support the creation, identification and commercialization of winning technologies and explain that this can occur in a market with autonomous such intermediaries, or by the same intermediaries being internalized within a business hierarchy as management functions. Competence Bloc theory is therefore a perfect theoretical platform to explain firm formation, and explicitly introduces such firm formation in MOSES (Eliasson, 2023b).

The business firm in MOSES is a financially defined entity. Its outer boundaries are determined in financial markets as different components (“divisions”, technologies) are acquired or spun off to reconfigure in new business entities. Since the dominant production capital in a firm is human based competence (The Firm as a Competent Team, Eliasson, 1990a), that is traded in the “labor market” (previous section) the financial and labor markets together form a powerful vehicle for building synergies or business hierarchical organizations called firms (Eliasson, 2021).

Ballot et al. (2006), and Eliasson (2023b), address the problem of the floating boundaries of business agents operating in realistically modelled markets in which financially defined business hierarchies (“business agents”) constantly break up and recombine or establish themselves as separate business entities. Competence bloc theory (see section 2) was formulated to explain that organizational dynamic. Ballot et al. (2006) explain, using a simple learning model, how financial intermediaries within a competence bloc may learn through investment experience to build the industrial competence needed to identify winning technological innovations, and support their commercialization and scale up to industrial levels. This venture into the markets for strategic acquisitions and divestments based on competence bloc theory offers promising future modelling experiments (Eliasson and Eliasson, 2005).

The MOSES model encapsulates the static general equilibrium model as a special case. The Capital asset pricing Model (CAPM) developed by Sharpe (1964), was derived from the general equilibrium model, that has nothing to say on an economy out of equilibrium, or how it may move from one equilibrium state to another. It therefore became interesting to study (through simulation analysis on MOSES) the reliability of the CAPM model as a predictor of asset prices in a model where such a market equilibrium did not exist, and the economy normally operated far away from anything close to a capital market equilibrium, the latter being a property MOSES philosophy equates with empirical relevance. Such a study became possible when a new capital market system was substituted for the industrial banking system of the original model in Eliasson (2001) and Eliasson and Taymaz (2001). Broström (2003) also demonstrated through simulation experiments on that general version that CAPM predictions were generally unreliable, and the more so the more out of “capital market clearing equilibrium” the MOSES model economy.

The rich specification of the artificial internal economy of the MOSES model can now be used in a new role as cognitive support in the choice of best partial model to study a particular decision problem, or to interpret the econometric results from an obviously miss specified partial model that can be derived as a special case of a MOSES type model. These applications are greatly facilitated by the modular design of the model. Since boundedly rational policy makers bulk so large in the economy they exercise a heavy leverage on the entire economy, not least when it comes to the biasing of price signals in markets caused by tax and income transfer wedges. More important are the potentially large consequences of policy mistakes due to lack of understanding of the economy wide consequences caused by their interventions. Economic modelers in their advisory business must therefore take model design and verification seriously, and the prior customization of the model’s specification in particular (Eliasson and Taymaz, 1992). MOSES thus also offers a way to add scientific precision to the final interpretation of econometric results that Feldstein’s advice lacked. Within its range of applicability the MOSES model therefore offers the possibility to derive useful partial models for econometric analysis that can be broadly interpreted with reference to the complete MOSES model system. Are you facing a situation of insufficient Keynesian demand? Does the economy lack the Schumpeterian entrepreneurs that maintain the diversity of structures needed for stable economic growth, or is market competition distorted by Government regulation such that market failures are in fact policy failures? Thus, MOSES allows the economic analyst to evaluate the social costs of being wrong in the choice of model, and hence of understanding.

The MOSES model has other advantages too. It not only takes the specification down to the agent level, uses empirically recognizable categories and emulates documented decision processes based on (statistical) information from Eliasson (1976a) to be used by firm management, and collected in the Planning Survey of the Federation of Swedish Industries, which was specially designed to set MOSES up empirically. Since the MOSES model is dynamically self-coordinated by market agents with (autonomous) price setting capacities, policy interventions can be expected to conflict with the agendas of agents (Eliasson and Taymaz, 1992) and thus interfere with this self-coordination.

The modular design of the core micro market selection based endogenous growth machinery of the MOSES model has remained the same since Eliasson, 1976b and Eliasson, (1976a). Its calibrated parameters have also remained reasonably stable over the decades of MOSES development (Section 3.9 above). In so far as any of the partial sub models of the MOSES model is taken seriously as an empirical representation of reality, any policy maker needing serious advice, therefore, should be more than happy to transfer his/her attention to the MOSES model.

I will therefore conclude this section with reporting on some of these odd cognitive applications of the MOSES model in the borderland between theoretical and empirical analysis.

It began already in the late 1970s when my department at the Federation of Swedish Industries got involved in an academic exchange of researchers with the then Soviet Union, notably the Central Economic and Mathematical Institute (CEMI) in Moscow. At the time the centrally planned Soviet economy was creaking in its joints. The CEMI had been charged with studying ways to introduce markets in the rigid Soviet economy and some visiting Soviet economists got wind of the MOSES modelling project, among them Nicolai Petrakov, one of the directors of CEMI. It was agreed that a compacted version of MOSES be transferred to a Norsk Data desk computer at CEMI to be loaded with Soviet data. I, then director of IUI, and Bo Carlsson, my deputy director, helped CEMI researchers in the 1980s to practice the transfer of a centrally planned system to a market economy on the MOSES model. The idea was good, but at the time MOSES was not ready for such advanced studies, and the compacted Soviet version of MOSES had very little left of the complex market dynamics in Eliasson (1976b).

Some academic papers on those MOSES applications on Soviet problems were however published by Soviet economists, but they were primarily devoted to presentations of the model to Soviet economists, and to identifying how a market economy distinguished itself (favorably) from a centrally planned economy, or for that reason also from the softer “dirigiste versions” practiced in France and Japan. Rather than how, we identified what not to do, and why static equilibrium modelling, needing the Walrasian central auctioneer to coordinate the economy, was a bad idea. Antonov and Trofimov (1993) demonstrated that mandating the firms in the MOSES model economy to follow Keynesian or Neoclassical macro model forecasts (reestimated every quarter on data generated by MOSES by the central planning authority) was a growth depressant. When freed of those mandates, the model economy outperformed the policy mandated version in generating very long run macro economic growth. When firms were left alone to use its own adaptive price anticipations and ad hoc decision rules, in the free market regime so defined individual firms happened to come upon, identify and commercialize winning technologies that they had missed under the constrained policy regime (Type II economic mistakes). (On the other hand, the GNP composition of output under the positive market scenario was endogenously determined and also beyond central policy control. In later MOSES studies we found that if global prices were developing close to a steady relative price change, and therefore predictable in expectation by firms’ adaptive expectations algorithms, a central policy aimed at technological imitation, rather than innovation, might be a medium term winner, however leading to a technological lock in into old structures and a dangerous competitive exposure to unexpected global technological innovation of the Internet revolution kind). In Eliasson (1998), finally the results of MOSES simulations were incorporated in summarizing the benefits of a free market economy, but also the problems of achieving an orderly transfer of a planned economy to a market economy. That, however, in turn required, it was observed, that the institutions supporting trade in intangible assets, notably property rights, be developed (Eliasson and Wihlborg, 2003). The property rights institution was one of the pillars of a market economy in serious disrepair in the Soviet economy.

In the early 1990s the Swedish economy found itself in a combined structural and financial crisis with stagnating production and severely reduced structural diversity. The Government asked Assar Lindbeck, a well known Swedish economist, to lead a Commission to suggest policy remedies. Special studies were subcontracted to economists across the country. I was asked to contribute something on micro to macro, a bit of Schumpeterian entrepreneurship and some results from our simulation studies on the role of human capital in economic growth (Eliasson, 1993). The Lindbeck commission came up with 113 detailed recommendations (Nya villkor för ekonomi och politik. Ekonomikommissionens förslag, SOU, 1993). It, however, failed to take the micro to macro dimension explicitly into account, and the deteriorating diversity of production structures that we pointed to appeared not to have been taken seriously. Most (not all) of the 113 explicit pieces of advice, however, would anyhow have come out of a MOSES based study. We therefore published our own IUI analysis of the economic situation in Andersson et al. (1993) in which the need for a market directed reorganization of production structures was argued to be needed to restore a precariously deteriorated diversity because of a long-term decline in entrepreneurial entry. This, again illustrates the Feldstein (1982) remark that all models are exactly wrong in some respects, but that they may still be used together with experience and common sense to come out vaguely right. It also illustrated the method discussed in Section 5.1 to use the MOSES economy wide and general framework as cognitive support for partial analysis. It would thus not have hurt the case to test at least some of the 113 policy propositions of the Lindbeck Commission on the MOSES model for the possibility of being wrong, or simply irrelevant in the wider perspective of a complex economic situation. It was there, ready to be used.³⁸

In his 2017 article Richiardi (2017) asks why agent-based modelling and microsimulation methods have experienced such difficulties of becoming a standard approach to macro economics. About these models Richiardi observes that; (1) ad hoc and not well grounded assumptions are often resorted to, (2) too many degrees of freedom mean models are non-falsifiable, (3) lack of empirical grounding and ad hoc calibration mean lack of empirical credibility, (4) poorly documented specification prevents replication of experiments, and (5) great demands on technical and programming skills make both the use, and expansion of agent based models difficult. They tend to exist for a while in isolation and then be forgotten. We have been aware of these problems throughout MOSES’ 50-year history, and they apply to almost all microsimulation models I know of, but this Anthology should show that they don’t apply to the MOSES project. This Anthology should rather make clear about 1,2 and 3 that there is little or no ad hoc economics associated with the MOSES project, and that we have taken validation of the model seriously, perhaps even too seriously. In fact (4) was easily overcome in Broström (2003), a master student at Linköping Technical University who set MOSES up for advanced simulation experiments and concluded his thesis in six months. Also (5) should not be that much of a problem. My 17 year old grandson managed to convert MOSES to object oriented code directly from the coding manual and without doing it by way of the current APL code, to run a few experiments, and to write Lindstenz (2023) in three weeks (Eliasson, 1976b³⁹ That the Planning Survey includes critical proprietary data is a problem that was overcome with the creation of a deidentified version for external users by Taymaz (1992a). To set MOSES up for serious empirical work on a different economy than the Swedish one of course requires a separate Planning Survey to establish an initial state for that economy. Some more fundamental obstacles must however be at work on preventing economists at large from taking advantage of the great potential of the microsimulation method.

There must also be something wrong with a profession that has not taken advantage of the potential of agent based economy wide modeling, its opportunities for relevant empirical applications and for using simulation mathematics for general economic theorizing, and it is interesting to ask why. Thus, the introductory two sections on general analysis of the artificial reality of the MOSES model of an experimentally organized economy.

I have made a point of the generality of the master versions of the MOSES model as they integrate a number of familiar partial models (Section 1.1), and the possibility to minimize the partiality problem by customizing the MOSES model from this master platform for particular applications (See section 3). Even so, technological and organizational changes in an advanced industrial economy like the Swedish one progresses at a rate that will leave the academic modeler constantly behind. And so do also the concerns of politicians wanting to change the composition of economic growth and the distribution of the benefits of economic progress. In order not to do worse they need the support of models capable of reliable advice on such matters. On this I have had the advantage of being for many years the head of a research institute (the IUI) that was constantly engaged in applied industrial economics research, and also been personally involved in down to the factory floor research (see Eliasson, 1976a; Eliasson, 1980b; Eliasson, 2013; Eliasson, 2017b). The MOSES specification was up to date on the structure of the Swedish economy during the early Post WWII period. Since then the real Swedish economy has been subjected to a number of fundamental structural changes, some of them being endogenously generated in the model, some attended to by a revised specification, and others only prepared for.

The most needed modifications of MOSES are:

1. Households are still represented in macro, as in a conventional eleven sector Leontief - Keynesian model. Converting the household sector onto a micro foundation has been prepared for empirically with the HUS project managed by Anders Klevmarken in cooperation with IUI (Eliasson, 1982; Eliasson, 1985; Eliasson and Klevmarken, 1981). A micro household empirical implementation can thus be naturally integrated with the 1982 database. It should also be noted that a satisfactory micro household implementation for the most interesting applications (for instance income and wealth distribution consequences of more or less successful growth scenarios) does not need a database of thousands of households, which should therefore not be attempted. A sample of households of different human capital characteristics should be sufficient. The most challenging task when it comes to individuals will be how to represent human capital relevantly, not least how human capital, the dominan production capital appears in the formation and operation of firms (Eliasson, 1990a; Eliasson, 1990b; Eliasson, 2023b). Micro specification of households should therefore make a more realistic representation of workers’ heterogeneous production capacities possible (see below and Section 4.4.2 above). The micro household extension of MOSES has however been held back by our early choice of APL as a computer language. As we have learned a transfer to object code, then not developed, has long been needed. Derviz (1992) prepared a conversion of MOSES to object code, but we have so far not found a budget to do that full scale, except that Lindstenz (2023) has converted the early Eliasson (1976b) MOSES version to object code and managed to get this version up and running.

2. An extension of great principal interest would be to establish a micro based intersection (market) between manufacturing firms and households. That extension is needed to explain the role of human capital in the creation and commercialization of new technologies , and its compensation in the form of capital gains through the spontaneous formation of new business entities , and how differences in market regimes have governed this most powerful income and wealth distribution widening process in recent decades. Steps in that direction have already been taken in that Ballot et al. (2006), republished in this Anthology, model how specialized knowledge intensive commercializing agents detach and merge spontaneously from manufacturing firms and households to attract financial resources and support macro progress. Ballot et al. (2006) limit MOSES simulations to the role of venture capital firms that learn from experience and eventually, if in sufficient numbers, and with the needed experience, influence the macro economy. The long time it takes for emerging macro consequences of such spontaneous agent formation to show was a particularly interesting result. The extension of the model to cover such spontaneous formation of specialized commercialization agents within competence blocs (Section 2.1) should be possible with the realization of §1.

3. The §2 extension is a special case of the outsourcing of specialized production and services, even critical technological and commercial ones, that has fundamentally reorganized the industrial landscape during the last two or three decades in advanced industrial economies, much of it being moved by the digitalization of the economy (Eliasson, 1996b; Eliasson, 2013). This in turn calls attention to the concept of the firm as a dynamic and internally constantly changing financially defined entity with unstable and endogenously changing intersections with its market environment. To stay relevant agent based economy wide modelling simply has to attend to this institutional dynamic and Eliasson (2023b) outlines a dynamic version of Coase’s theory of the firm, suitable for modifying the current financially defined MOSES firm in this direction.

4. Human capital is the dominant production capital of the advanced industrial economy. It is largely tacit and cannot be meaningfully represented, neither as the measured capital stock of neoclassical production theory, something business firms have understood all the time, nor as a top-down managed hierarchy. Tacit human competences are decentrally allocated within hierarchies and over markets for competence, sometimes called labor markets, to form competent teams, called firms (see Eliasson, 1990aEliasson, 1990b). The great challenge would be to model the creation of new tacit knowledge and its endogenous agglomeration in markets into firms or competent teams . A revised firm model in MOSES that captures this has long been prepared (see Eliasson, 2021; Eliasson, 2023a; Eliasson, 2023b, and could naturally be integrated with the work on labor markets and human capital accumulation in the MOSES model of Ballot & Taymaz (see Section 4.4).

5. Introducing human capital explicitly in MOSES however raises the difficult problem of modelling quality. To some extent Gerard Ballot and Erol Taymaz has already made human capital heterogeneous in MOSES (see Section 4.4.2), but a micro household sector (§1) would be a great step forward in that respect. §4 above is about how markets evaluate human capital in production decisions. More problematic will be to represent quality differentiation in product markets, especially since new technology creation over MOSES history has shifted from allocating R&D investments to the raising of manufacturing productivity towards increasing product qualities. The assumption that markets perfectly represent product quality change in prices was reasonable before the new Millenium, but is no longer. With households being modelled in micro, not only labor can be made heterogenous, but also the possibility of modelling customer competence in appreciating product quality in their purchasing decisions.

6. As I have suggested several times, the MOSES model is an excellent platform for formulating a general, market based theory of industrial economics, as distinct from the static industrial organization (IO) models that lack general markets that connect agents to macro (Ballot et al., 2015). Microsimulation methodology is the appropriate tool to advance such new theory formulation. On this I agree with Richiardi et al. (2024) that micro simulation and agent based modelling are just versions of the same thing, and that the two ought to have fruitfully converged long ago. Tunnel perspectives on both sides have prevented learning from one another.

7. An interesting study of the kind I am advocating would be to address the ergodicity characteristics of dynamic economy wide modeling, that Samuelson (1968) argued are fundamental to good economic modeling, in the context of the transactions costs intensive market self-coordination of the MOSES model economy. Ergodicity is a special case of self-coordination, namely when markets guide the economy to the same final state, irrespective of initial conditions. Since the models Samuelson was referring to have no markets, my proposition is theoretically different. A wider definition of ergodicity, applicable to the MOSES model economy would therefore be to study under what market circumstances the MOSES economy stays bounded from above and below under a market self-coordinated regime, and to investigate what happens when that wide band within which the economy evolves is gradually narrowed by policy intervention to, for instance, make markets more efficient, or to remove cyclical unemployment spells (for such countercyclical policy experiments on MOSES (see Eliasson and Taymaz, 1992 in this Anthology). The more narrowly bounded the closer to the mathematical definition of an ergodic model it would come. I have argued that an interesting policy experiment would be to try to” policy manage” the MOSES model economy into a reasonably narrow bounded such range that expands as rapidly as is possible without becoming unstable over the longer run. Trying to pin the MOSES model economy down on a steady state, and the same steady state irrespective of initial conditions, the ultimate definition of an ergodic process, we already know will be a vain ambition, since such a state is not an attainable state in the MOSES economy. However, trying to identify a maximum possible welfare compatible such expansion range through policy simulation would be an interesting policy study (see section 1.3).

   The MOSES economy is of course not free of external constraints. Its self-coordination property for most of its life has occurred under an upper exogenous global technological boundary, a constraint that was partially lifted with the Ballot and Taymaz (1998) firm based endogenous innovation module. Exogenous foreign prices, including the foreign interest rate still constrain market self-coordination, even though there is no lower boundary, since the model economy might collapse completely if development reduces its diversity of structures too much. For instance, markets in MOSES stop functioning when the number of firms has been reduced below three. They stop functioning well already when agents become too similar (too “representative”, Eliasson, 1984; Eliasson, 1991b; Eliasson, 1991a; Eliasson, 1991c), and there is no Walrasian central price setter in MOSES capable of taking over the market coordination job. A study of fundamental theoretical interest would therefore be (see Section 3.4) to explore the market self-coordination, and thus “ergodic”, properties of the MOSES economy further by endogenizing also global price determination and the interest rate. The Ballot and Taymaz (2012) cloning of MOSES models into a global economy has already laid the foundation for such a study. By adding trade in innovations (technologies) between the cloned MOSES economies, the global pool of best practice technologies could also be endogenized, and foreign prices be endogenously set by trade between the cloned economies. Such experiments on the MOSES model economy would take us closer to understanding the dynamics of an Experimentally Organized Economy.

To be remembered is that none of the above suggested expansions of the MOSES model will need full scale empirically implemented and recalibrated versions. The fact that the MOSES project has been prepared for a full scale empirical implementation that has not yet been achieved has a historical explanation. Something such ambitious was a prerequisite to get the project started in late 1974, and it is still within a realistic horizon. The generalization work sketched above can however take place more modestly and independently of realizing the full-scale empirical ambition, and on the basis of what has already been done. But the theoretical (“academic”) studies on MOSES that we currently focus on will make the model better prepared for future empirical projects, and with time such a full scale empirical inquiry will be necessary to build a new structurally upgraded empirical master platform for future theoretical analyses.