The limits of Policymaking - An analysis of the consequences of boundedly rational government using the Swedish Micro-to-Macro model (MOSES)

In an often referred to paper Lucas (1976) argues that government should be very restrictive in attempting to correct undesired developments in the economy (fine tuning) to make it possible for agents to form rational expectations (RE).

Rational expectations in Lucas’ and our interpretation are such that agents will (eventually) learn to be right in expectation. Studies addressing the problem of RE have all been quiet on both the time dimension of the “rational learning” process, and the assumption of zero or negligible learning costs, such that these transactions costs do not influence the position of the exogenous equilibrium of the model assumed to exist and to be rationally aimed for. Furthermore, the possible inconsistencies of policy ambitions and agent preferences are generally neglected.

We study a micro (agent) based macro simulation model of a Swedish like economy in which heterogenous firms form expectations about their market environment, which consists of all other firms in the market, and are affected by policies.¹ Firms’ learning is costly, and costs mostly come in the form of expectational mistakes (all other costs being regarded as production costs). Such business mistakes are agents’ exits, mistaken investments, production planning and hiring decisions that generate below targeted profits. We observe already here that such business mistakes in the model are non-stochastic (deterministic) events that make rational expectations non attainable. The reason is that agents’ expectations and planning mistakes feed back on firms’ production planning, recruiting and investments from period to period, affect production structures of the model economy and turn it into a non-stationary process which is incompatible with RE. We will inquire into the time dimension of the learning processes of business agents, including Government to study what happens when agents strive ambitiously to attain ex ante perceived equilibria that “vanish ex post” because of the constantly ongoing market induced structural changes, coming faster and being more disruptive the closer to the perceived ex ante equilibria competition pushes the economy.

The Swedish MM model approximates what we call (Eliasson, 1987; Eliasson, 1991a) an Experimentally Organized Economy (EOE). The behavior of such a model economy, because of the underlying non-stationarity of economic processes, will at each point in time represent an unpredictable economic environment for each agent operating in its markets in the sense that expectations and plans will not be right “in expectation”. Hence, agents will commit systematic expectational mistakes. For each agent (firm) this means that it becomes rational to learn how to cope with mistakes, rather than to try to learn to improve forecasts. Even if the agent varies its business decisions repeatedly, they will never be right in expectation (Eliasson, 1990a; Eliasson, 1990b). The situation will be the same for the policy maker or government, which faces the additional problem that its actions significantly affect the actions of all agents and thereby the entire economy. When agents are not passive recipients of Government policy, but play games with Government, which then in turn play games with agents, the whole idea of a better informed, or a fully informed central Government with perfect overview, will have to be abandoned. Central Government rather runs the risk of making things worse, trying to beat the market, with large and even permanent negative consequences for the entire economy. The conclusion is, hence, that rather than collect more data, to learn to predict better, it should do less.

Behind these propositions lies an intriguing theoretical, perhaps even philosophical problem. One might argue that if the Government knows the specification of the model, including the behavioral characteristics of all agents, which are deterministic algorithms, it should be able to predict the outcomes of all its interventions, including the reactions of agents. In principle perhaps. In the reality of the model not so. We know the model exactly (the ”truth”), but the Government or policy maker in the model does not. It will have to learn like all other agents, and learning the specification of the economy wide dynamics of the MOSES model economy through repeated studies of the data it generates, including its responses to its own interferences, however many and heavy interferences it enacts to learn, is not theoretically possible, and in practice a vain pursuit. Hence, the inability of agents to form RE depends on fundamental market mechanisms at work even if Government stays entirely awa.

The Experimentally Organized Economy (EOE), in which our policy experiments will be run, is represented by the Swedish agent based Micro-to- Macro (MM) model MOSES (see next section). The MOSES model features endogenous economic growth under an upper (exogenous) global technology constraint. The rate of growth achieved under that constraint depends on agents’ competitive performance in product, financial and labor markets. This model allows us to investigate several problems not tractable by received macro and static general equilibrium models.

1. Selection: Phenomena such as entry and exit, mergers & acquisitions, and Schumpeterian creative destruction arguably have long run economy wide consequences that cannot be neglected in studies of economic growth. The consequent non linearities and initial state dependencies make static equilibrium models mistaken tools and macro models useless. The MOSES model features competitive market selection of agents and endogenous populations of firms.

2. Bounded rationality: Agents in the MOSES model are boundedly rational in the sense of Simon (1959), and tacit knowledge in the sense of limited communicability is a consequent characteristic of agent’s knowledge capital (Eliasson, 1990a; Eliasson, 1990b). With all agents, including Government, being boundedly rational, there will be no such thing as an information advantage of the central policy maker, as is conventionally assumed in standard equilibrium analysis.

3. Heterogenenity: Agents in the MOSES model are heterogenous and differently sized statistical representations of real Swedish firms that compete for customers in product markets, for workers in the labor market and for resources in the financial market. Doing away with the standard assumption of received economic theory of representative agents of atomic size allows us to talk about a market economy self-coordinated by boundedly rational, or even ignorant agents.

4. Stockholm School expectational dynamics: Agents form expectations about (for instance) prices using adaptive learning and error correction algorithms (Eliasson, 1976b; Eliasson, 1977; Eliasson, 1978a; Eliasson, 1980). Period to period feed backs of expectational mistakes at agent level cause genuine market dynamics. Decision mistakes affect long-term economy growth and is regarded as a normal (transactions) cost for economic development.

5. Dynamic market integration: Product, financial and labor market considerations are integrated in the business plans of the boundedly rational firms in the MOSES model economy. Partial analysis becomes a misspecification.

6. Intergenerational dynamics: Since agent mistakes, including Government policy mistakes, cause socioeconomic costs in terms of less than feasible very long term macro-economic growth, ambitious policy fumbling based on misunderstanding economy wide dynamics is therefore likely to incur long term, perhaps permanent costs to future generations.

7. Agents playing games against the policy maker: Standard analysis assumes agents passively adjusting to central policy directives. Firms and households may however understand that their agendas do not tally with that of central Government and attempt to avoid being cheated upon. In our case firms begin to play games against the Ministry of Finance. Sustainability of policy is defined by the time it takes for agents to learn to play successfully against Government. When policies are frustrated, Government may respond by suddenly changing the rules of the game, for instance instituting a tax reform. Increased such political uncertainty, however, spills over into market uncertainty and causes economy wide underperformance. Some simulation experiments have been designed to illustrate the outcomes of such policy gaming.

8. International integration: Most policy models view the national economy as a controlled system with well defined (statistical) limits. International economic integration of markets has not only limited the scope for independent national policy action. It has also made the statistical limits of the national economy increasingly diffuse, further reducing the information content of national statistics and the central controllability of the economy. Global financial markets have therefore become an enemy of national politicians. The policy maker must understand the consequences of endogenous international movements of capital on the national economy to exercise national policy control. Price sensitive exports and imports are complicating factors, as are interest sensitive credit trade flows, both part of the MOSES model.

This essay addresses the possibilities for a myopic Government to control national unemployment short term in an internationally integrated economy with individual agents with own agendas and independent decision leverage, without shifting the costs of that policy in the form of less growth and larger unemployment onto future generations.

In modelling market information processing and policy making of government relevantly we must place economic agents and Government in an explicitly modelled market environment in which their own instruments of learning about, and interpreting that environment not only are explicit but also allowed to influence the environment itself. In this we depart from the assumptions of currently popular and rational expectations inspired statistical learning theory (see Lindh, 1993). We have found it particularly important not to use - as in statistical learning theory – an equilibrium model with costless learning of agents, which by assumption remove the important and difficult problems that face economic decision makers, notably the risk of making systematic decision mistakes, the consequences of which may cumulate. In setting up such an unconventional decision stage for government the possible existence of a rational expectations equilibrium is considered. The Swedish Micro- to - Macro model MOSES exhibits the properties that we consider characteristic of an Experimentally Organized Economic environment, and hence of a real economy (Eliasson, 1987; Eliasson, 1991a).

Different experimental designs usually exhibit small differences in the medium five year or so runs, except for certain parameter settings that change cyclical frequency. Only in the very long (beyond 15-20 years) run cumulative development differences begin to show at the economy wide level, exhibiting both the path dependence of the MOSES model economy, and a degree of robustness that protects it from external disturbances. That robustness depends positively on a combined diversity (of structures) and market flexibility that are associated with an advanced industrial economy. Market disorder and collapse like behavior at the macro level may occur in the very long run for certain parameter settings, saying that market self-coordination doesn’t work well for all parameter settings and forever. It therefore becomes theoretically and principally interesting to discuss how to interpret the empirical relevance of this property.⁶ In fact, experiments on the MOSES model are the perfect instrumentation for such analysis. The point has been made already in other papers (e.g., Eliasson, 1983; Eliasson, 1991a) that the MOSES model economy, while not sufficiently diverse to stay dynamically robust forever, or for as long and for the kind of exogenous and policy damage the Swedish economy has been subjected to over the years they demonstrate the importance for policy makers, to build structural robustness into their economies.

4.1. One percent unemployment target – simple sequential policy rules (SIMP1)

On average the government hits its target accurately. Compared with the BASE run in Table 1 Government sets an ambitious unemployment target of 1 percent compared with no target in the BASE run and uses its forecasts to determine how much Government recruitment is needed for that. Manufacturing firms absorb the large increase in wages occasioned by the increased demand for workers (Figure 1) by reducing slack and raising productivity (Figure 2) and, when this no longer works, by exiting. The manufacturing industry gradually becomes more concentrated and less diversified, but negative effects take a very long time to show at the economy wide level. After 15 years manufacturing output stagnates and then drops suddenly not to recover during the simulation period (Figure 3). Towards the end of the period 98 percent of the labor force is employed by the public sector that accounts for 66 percent of GNP by standard national accounting principles. Unemployment is kept roughly at the targeted 1 percent level (compared to a long run level of around 4 percent in the calibrated BASE reference case (Figure 3) which would be considered a rather healthy level for normal transactions turnover of labour), but instabilities in market prices begin to cumulate after 15 years or so. The public deficit explodes (Figure 4), as does the trade deficit, and the domestic interest shoots up. Obviously, this is not a sustainable policy scenario. If the government had not been intending to transfer all production to the public sector it would have called off the policy experiment at the latest after some ten years in order not to cause large cumulating welfare losses for later generations. The problem is that using its crude macroeconomic models and watching aggregate data, signals of an impending collapse do not show clearly until after some 15 years. So careless policy makers and inept economists/advisers may cause long-term disaster in the MOSES model economy.

Table 1

	ZEROBASE	EQBASE	BASE	SIMP1	SIMP2	KEY2	RAT2
QTOP*	150281	152394	153153	156675	134675	146823	141347
TEC*	11662	11625	11917	11981	11536	11902	12052
Labor prod.*	389439	385954	394917	797181	520333	659736	547376
Average annual growth rates
GNP	3.77	4.06	4.11	5.37	4.41	5.04	4.57
	0.74	0.69	0.75	0.72	0.58	0.83	0.67
Output	5.04	5.03	5.37	4.70	4.76	5.12	5.27
	1.34	1.54	1.40	2.12	1.57	1.80	1.66
Sales	11.23	11.26	11.57	10.96	11.00	11.36	11.49
	1.15	1.40	1.42	1.87	1.51	1.68	1.65
Prices	5.87	5.88	5.88	5.87	5.91	5.87	5.91
	0.53	0.53	0.52	0.54	0.55	0.54	0.54
Wages	6.57	7.85	7.93	14.04	10.28	12.46	10.56
	0.35	1.01	1.36	2.26	0.99	2.10	0.95
Profit margin	45.06	42.70	43.09	37.20	39.66	37.38	39.52
	6.16	3.79	3.75	1.72	2.06	1.37	2.44
Unemployment rate	8.04	5.19	5.28	1.06	2.24	1.85	2.14
	0.95	1.92	2.07	0.19	0.27	0.90	0.32
Exchange rate*	1.59	1.72	1.71	2.03	2.03	2.03	2.03
Interest rate	12.63	12.68	12.59	10.28	12.05	10.83	11.89
	0.87	0.57	0.56	1.26	0.69	1.74	0.95
Public deficit/income	-28.33	-29.96	-29.77	-42.48	-36.71	-40.56	-36.62
	6.18	3.93	4.24	4.34	0.84	1.90	0.80
Gov. share in GNP	35.33	36.29	36.22	44.75	40.99	44.00	41.02
	0.57	0.39	0.38	5.08	2.71	3.75	2.52
Trade def./GNP ratio	4.31	5.19	5.12	15.00	10.63	14.43	10.85
	2.23	1.27	1.38	6.02	2.62	4.57	2.63

1. Note: The standard deviation of a variable over the period is shown in the second row.

2. *

   denotes the value at the end of the period.

Figure 1

Download asset Open asset

Figure 2

Download asset Open asset

Figure 3

Download asset Open asset

Figure 4

Download asset Open asset

The seemingly positive “first period” policy results on GNP and manufacturing productivity growth apparently have been achieved by excessive public recruitment. But the explanation is not typically Keynesian. The more job openings in the labor market, the less risky it is for individuals to quit jobs and to take their time looking for a new job. Workers will increasingly (because of policies) go to the shrinking share of very productive manufacturing firms capable of paying more,⁷ or to Government. Low productivity firms are forced to contract production or to exit. Average productivity in manufacturing temporarily increases because of an improved static allocation of workers. Manufacturing output temporarily increases, but not much because of the growing exit rate.⁸ This is not a typical Keynesian demand effect, but it can be triggered by Keynesian demand policies. We recognize these types of market induced mechanisms on the MOSES model economy from several earlier studies (Eliasson, 1978b; Eliasson, 1978c; Eliasson, 1983). Provided demand policies are not excessive, as they become with time in the SIMP1 policy scenario and provided industry innovation and entrepreneurial entry are sufficiently forceful to maintain diversity of production structures and heterogeneity among business agents this policy works (see next policy experiments and graphs). If pushed too hard, however, this policy causes cost overshooting and disorderly markets and eventually destabilizes the entire economic system.

While in the BASE case the economy trots on nicely for the entire simulation, and the government deficit even begins to decrease, unemployment levels stay at roughly 4 percent, and perhaps somewhat higher than a healthy natural turnover of workers would generate. The SIMP1 scenario eventually takes the entire economy close to collapse (Figure 3) with an abnormally low unemployment rate and almost all workers employed by the public sector (Table 2).

Table 2

	ZEROBASE	EQBASE	BASE	SIMP1	SIMP2	KEY2	RAT2
QTOP*	454640	391980	408493	41397	159049	53993	222153
TEC*	17992	19155	18840	22463	16143	19934	17895
Labor prod.*	722351	719729	729879	788204	825031	1011372	922739
Average growth rates
GNP	3.77	4.06	4.11	5.37	4.41	5.04	4.57
	0.74	0.69	0.75	0.72	0.58	0.83	0.67
Output	3.11	3.29	2.83	..	1.80	2.80	1.16
	3.09	2.28	2.40	..	1.36	...	2.33
Sales	9.36	9.66	9.21	7.23	8.52	11.99	7.79
	2.25	2.05	2.07	23.12	1.79	16.26	2.13
Prices	5.87	5.99	6.01	8.00	6.54	6.81	6.49
	0.62	0.57	0.58	3.59	0.86	2.60	0.95
Wages	5.14	7.25	7.71	11.75	9.34	9.63	8.93
	0.58	0.82	0.64	1.45	0.20	1.23	1.01
Profit margin	70.69	57.07	56.54	8.48	50.29	31.01	54.28
	6.58	5.60	4.76	67.57	4.09	49.60	4.30
Unemployment rate	13.46	4.60	4.13	0.98	2.08	1.91	2.02
	1.31	0.70	0.54	0.43	0.20	1.07	0.24
Exchange rate*	1.17	2.06	2.20	3.73	3.73	3.72	3.72
Interest rate	8.59	12.68	13.37	22.33	21.60	23.33	21.95
	2.52	0.40	0.93	3.72	4.18	2.79	3.94
Public deficit/	-2.21	-19.19	-21.92	-60.50	-41.86	-55.78	-43.86
income	7.15	3.06	0.97	5.18	2.21	5.30	2.21
Gov. share in GNP	35.28	39.09	39.91	66.01	52.41	58.90	51.69
	0.68	1.07	1.65	13.85	4.07	4.92	3.57
Trade deficit/	-4.03	2.43	3.32	47.05	23.94	32.74	21.43
GNP ratio	2.29	0.87	1.04	28.93	6.05	7.62	4.77
Public emp*	44.75	54.52	56.61	98.24	78.06	92.61	80.82
Private emp.*	41.01	39.45	38.44	1.01	20.33	5.69	17.75
N new firms*	72	54	57	24	52	37	51
N nullified firms*	42	50	52	176	69	154	80

1. Note: The standard deviation of a variable over the period is shown in the second row.

2. Labor force at the end of simulation is about 2530000.

3. *

   denotes the value at the end of the period.

Beyond the estimation problems associated with selection and highly nonlinear models (that we don’t discuss here) the above simulation experiments have some interesting methodological and theoretical implications. In an Experimentally Organized Economy (EOE), as represented by the MOSES model economy, endogenous entrepreneurial entry, and learning by, and competition among boundedly rational agents propel economy wide evolution. In a positive scenario this means a flow of successful growth promoting business experiments followed by a smaller but still significant flow of mistaken business experiments. Competition forces low performing firms to slow down or exit. If not countered by investment in high performing firms and entry, firms tend to become structurally more alike and markets more concentrated. The economy will then eventually begin to exhibit signs of instability. A balanced turnover of firms is important for the diverse production structures associated with a robust economy to be maintained (Eliasson, 1978b; Eliasson, 1978c; Eliasson, 1983; Eliasson, 1984; Eliasson, 1991a; Eliasson, 1991b). Table 2 shows that in the extreme policy cases SIMP1, SIMP2, KEY2 and RAT2 this balance has been disturbed in that more firms have exited than entered.

One sign of increasing instabilities in the economy is the development of disorderly markets that lower the information content of current prices as predictors of future prices that raises the incidence of business mistakes and impairs the allocation of resources.

Similarly, diversity of production structures can be damaged by ambitious stabilization policies aimed at reducing slack in the economy by competing high-cost producers out of business. In Eliasson (1983) it was suggested that increasingly ambitious Keynesian stabilization policies during the post war period indeed increased economic growth for several decades, eventually to reduce diversity of structures to the extent that the economy became precariously sensitive to disturbances of the oil crisis type of the 1970s. And the unbalanced ambitious employment scenarios, notably SIMP1 do not contradict that. proposition.

The policy experiments of this essay have been designed to emulate those policies and have done so quite successfully. The diversity of structures in terms of rate of return distributions over the firm population are shown in Figure 5 for the BASE, the SIMP2 and the ZEROBASE policy scenarios. This time data for a longer 50-year period have been studied, and rate of return distributions printed out for years 1986, 2016 and 2036.

Figure 5

Download asset Open asset

The SIMP1 policy experiment reduces the number of firms to the extent that the economy is close to collapse by year 50 (Table 2). The more moderate SIMP2 policy experiment is not that destructive, even though more firms exit than enter and the private manufacturing sector is left dramatically reduced compared to the public sector on the 50-year horizon. The distributional characteristics in Figure 5c show an unhealthy development of deteriorating diversity. On the horizon 2036 a few highly productive and profitable firms account for most of manufacturing production. This is a precarious structural situation. The opposite case (ZEROBASE) is shown in Figure 5a, with a healthy spread of rates of return across the large part of population of firms on the horizon, perhaps “too healthy” in that rates of return for most firms range too high above the interest rate in Figure 6, indicating a lack of competition. Here the BASE case shows (in Figure 5b) a distributional picture the way it should look in a robustly structured and competitive economy. Macro performance in terms of output and productivity growth in the BASE and ZEROBASE scenarios are roughly similar. A less ambitious public sector hiring policy has, however, kept wages lower and left firms better off in the ZEROBASE compared to the BASE case, and perhaps too well off because of lack of competition. As suggested earlier, optimum sustainable policy appears to lie between ZEROBASE and BASE, and close to BASE.

Figure 6

Download asset Open asset

Competition forces firms to become more productive, which is desirable even though it is achieved through shedding workers, but not if engineered though excessive public sector growth, and it is especially bad if public sector service production is not preferred by households over manufacturing goods.

Pushing the market-competition exposed manufacturing industry harder, makes wage and profitability distributions flatter, as in the ambitious policy experiments, and pushes economic structures closer to a state it is supposed to look like in static market clearing equilibrium. Even though neo walrasian economics wants it there, in a dynamic market economy which requires a certain amount of slack to be stable, apparently this is not a desirable state to aim policies for (Eliasson, 1983; Eliasson, 1984; Eliasson, 1985; Eliasson, 1988; Eliasson, 1990a).

Some say that it does not make sense to simulate an economic model for 50 years. We know for instance that weather forecasts beyond six or seven days are more or less without “weather content”, since the resolution and quality of initial state measurements are such that the precision of calculations deteriorate quickly as initial state and rounding errors cumulate in path dependent dynamic models.

This is of course true also of economic models when the accuracy of forecasts matters, or when the precision of effect or cost benefit studies is a concern. Ours is, however, a principally different problem. We primarily want to study the properties of a numerically specified model of a “Sweden like industrial economy”, and only secondly to calculate the magnitudes involved. That is analogous to demonstrating the existence of certain properties of a theoretical model. If it can be demonstrated that the precision of initial state measurements and of estimated model parameters do not matter for a particular result we have reason to be satisfied, and we might as well run the simulations for very long to see if the results show tendencies to cumulate. First, therefore we have the question of whether the demonstration of existence is economically credible, second whether the results carry over to the Swedish economy.

There are strong reasons for taking the economic mechanisms that have caused the long run collapse like simulation outcomes of this essay seriously as empirical economic phenomena. They are the results of well-known economic principles integrated in an empirical market framework of initial state measurements and parameters calibrated on empirical data. In that sense the analytical (simulation) outcomes carry more empirical content and credibility than pure theoretical reasoning within, for instance a computable general equilibrium model which is wrongly specified for such analysis. In principle the collapse like behavior of the economy caused by long run policy mistreatment should therefore be considered well established. Whether a policy action between BASE and ZEROBASE would have been optimal for Sweden is however a more controversial statement. The latter essentially means that the Government should have expanded less, albeit not altogether, and that it is now high time to reverse policies to avoid bad economy wide future outcomes.

Long run simulations also raise the problem of robustness of the model economy. Our experience from simulation analysis is that the MOSES model is a structurally robust economic system, and the main reason is that endogenous entrepreneurial entry in calibrated model versions (including the one used in this essay) can maintain a healthy diversity of production structures over the longer term. The consequence observed in the policy simulations of this paper is that very heavy policy mistreatment, such as SIMP1 is needed to force the model economy into a collapse prone state. This outcome furthermore takes a long time to materialize, well into the second period in Table 2 or after two decades or more. The typical characterization of the deteriorating robustness is the flattening of Salter curve distributions (in Figure 5) and a diminishing number of market agents (Table 2), both circumstances causing markets, and market self-coordination of the economy to lose functionality.

There are many other reasons for being interested in long run historic simulations of the MOSES model, for instance to investigate the theoretical meaning of a quantitative model of economic growth. We argue (Eliasson, 1991a; Eliasson, 1991b) that for a growth model to make sense it needs to be capable of clarifying whether economy wide endogenous market self-coordination supported by agent competition in explicit markets can sustain long term growth in the macro economy. Is the “invisible market hand” of Adam Smith sufficient for that, or is there a need for the “visible hand” of Walras’ central auctioneer to enter to correct economic development because of market self-correction failure. Neoclassical or neo walrasian equilibrium models assume the latter. In them markets have been eliminated and macroeconomic growth made exogenous. At the 1983 conference organized by the IUI and the Journal of Economic Behaviour and Organization (JEBO) in Stockholm and published as Day and Eliasson (1986), we asked wether markets needed a "boss" to function properly, and answers came out between yes and no depending on which model was guiding the discussant . The MOSES modelling project has from the start been focused on the relative importance for economic evolution of the invisible hand of dynamic markets and the visible hand of a central policy maker (Eliasson, 1976b), and in this paper we have been able to demonstrate that in a market self-coordinated economy wide model the visible central policy hand of Walras is often likely to mess up the market self-coordinating machinery because of a bad understanding of the dynamics of a market economy. The policy maker should therefore not only take it easy but in general avoid ambitious policy experiments because of the heavy leverage they have on the economy.

We also dare to generalize this result to any economy in which government distributes a significant part of its resources through the public sector, even though this may look like a contradiction. Why shouldn’t Government then rather use a MOSES type model to control the economy through varying its parameters. Then policies can continue to be ambitious and also be more informed.

Not so! This is where the objections to very long run forecasts, or forecasts in general, enter. Like any model the MOSES model will also generate error prone forecasts of the real economy it represents because its specification, however detailed and comprehensive, will always be partial. The simulations we have presented are not forecasts of what will happen, they are predictions of what will happened in controlled policy experiments on a model economy, or what is likely to (approximately) happen over the longer term in the real economy that the model has been designed to represent. The MOSES model, however, gives a more varied story of what might happen, side effects and all, than most forecasting models used in policy practice are capable of telling.

The above document is a sligthly revised version of the IUI working paper from 1992 with the same title that included a whole lot of handwritten corrections made as the paper was prepared for publication (see fotnote 1). We considered this publication important because its results became an input in the IUI Long Term Survey referred to as Andersson et al. (1993), notably chapter 7 with the title. The Limits and possibilities of policy making, and greatly encouraged us to voice strong recommendations to dramatically reverse current policies, and to stop preventing unemployment from increasing by holding back a long overdue reallocation of labor from crisis (or more or less so) industries to expanding businesses.

One complication held us back. In early 1993 (dated February 8) one of us (Gunnar Eliasson) received a letter of positive criticism of the original working paper from Professor Odd Gulbrandsen, an old IUI associate, with experience from large scale modelbuilding at IIASA, and just retired head of the research division of UNCTAD, who had read the paper carefully. Professor Gulbrandsen brought up the principal problem of extreme “predictions” associated with the longer term dynamic simulations of complex models that were always partial however well founded. Since the original working paper of this study is of interest in itself because it became a direct input in the IUI Long Term survey (Andersson et al., 1993), and there came out in the form of strong recommendations to reduce ambitious unemployment policies and make sure that workers were transferred to growing firms from crisis industries, even at the cost of a higher transactions unemployment. Gulbrandsen’s critique therefore also relates to the way our simulation results were used, even though the Andersson et al. (1993) study had not been published when he wrote his letter. It rather points to the general methodological problem of the empirical relevance associated with large scale initial state dependent and difficult to estimate microsimulation models like MOSES. We therefore chose to complement the revised document with this methodological postscript, and in particular comment on the relevance of extreme model predictions of the outcomes of extreme policies.

The MOSES model runs on a rich statistical input, mostly in the form of initial state measurements based on a specially designed so called Planning Survey conducted annually by the Federation of Swedish Industries and covering all large Swedish manufacturing firms plus a sample of small and medium sized firms. Those initial inputs are then all reproduced dynamically from quarter to quarter throughout each simulation and can be studied and tabulated at any time during a simulation. Tables 1 and 2 contain only a fraction of the data output the model can generate. Not rarely have we called up simulated additional data to figure out what has happened during an experiment.

The problem is that the MOSES model is calibrated on only a fraction of that data by manipulating an equal number of parameters. All other parameters of the model, including direct measurements of for instance the initial state (a complete stock and flow consistent micro to macro data base was brought in directly from external sources (see Albrecht et al., 1992). In the beginning calibration was on a selection of targeted macro time series as in Eliasson and Olavi (1978), but later with the computer-based calibration program of Taymaz (1991) on a much wider spectrum of target variables, also including distributions of micro data (see for instance Albrecht et al., 1992, and Eliasson, 1985). In doing so we tried to keep the number of manipulated parameters equal to the number of targeted variables. The targeted variables were however still only a fraction of MOSES standard simulation output, and of what it could deliver in the form of detail.

The target variables for calibration were chosen to be those that we had an ambition to explain endogenously, namely (1) critical economy wide growth variables, (2) macro cyclical variables, some of them the same as under (1), including also important price indexes, and later (3) some distributional characteristics expressed as the Salter curves in Figure 5. Sometimes, therefore variables not targeted for calibration exhibited unexpected data constellations, sometimes fine details that we could explain in terms of their generation in the model, but that were far from what you would expect in reality. Firms, even flagships of Swedish manufacturing industry could go bankrupt and exit. Since we had no ambition to predict correctly at that level of fine detail it was not documented, in this case for confidentiality reason. But sometimes aggregate output was tabulated and observed by readers and reruns of entire experiments necessary to explain them technically. Professor Gulbrandsen brought up some odd numbers in his letter.

The interesting counterquestion now is: Are these outputs empirically reasonable. In a carefully specified and perfectly estimated model you would be inclined to say yes. Now MOSES’ parameters were only calibrated, so one of them being very wrong could explain the outcome. The main problem in this case, however, is that the difficult to explain model predictions came from extreme policy experiments, such as SIMP1. Even though Swedish policy making had been extreme during the 1990s it was never that extreme, and the policy ambitions of SIMP1 would most probably have been abandoned long before the end of the simulation. So, carrying it through with years annotated and all would produce a lot of difficult to explain detail.

Should such odd predictive output that we don’t recognize as empirically possible, being the outcome of extreme circumstances be documented or not? If you believe in the model, they could occur under unlikely circumstances. Even the most incompetent policy maker would not run the economy to the ground as in Figure 4. Some odd constellations of predictive output that even the modelers would not discover before printing the table are likely. And when discovered and technically explained they may be found to be of methodological interest, or perhaps even a relevant possibility to consider. Above all, they are properties of the model economy that we study. So of course, we should document all data and wait for observant criticism, and then decide whether we are arguing about a real economic phenomenon, or an abstract theoretical possibility. This is not different from drawing conclusions from preliminary data that may have to be completely revised when new data arrives.

There is finally no way to avoid priori assumptions to appear as part of empirical explanations. Keynesian macro model advice on policy is full of such priors. That extreme policy making will generate extreme outcomes should surprise nobody. As a rule, however, extreme policies are terminated when the damage becomes statistically visible.

Figure 7

Download asset Open asset

1. 1
   MOSES Code Research Report no.36

   1. JW Albrecht
   2. F Bergholm
   3. G Eliasson
   4. KA Hanson
   5. C Hartler
   6. M Heiman
   7. T Lindberg
   8. G Olavi

   (1989)

   MOSES Code Research Report no.36, Stockholm:IUI.

   • Google Scholar
2. 2
   MOSES Database, Research Report No. 40

   1. JW Albrecht
   2. P Braunerhjelm
   3. G Eliasson
   4. J Nilsson
   5. T Nordström
   6. E Taymaz

   (1992)

   MOSES Database, Research Report No. 40, Stockholm: IUI.

   • Google Scholar
3. 3
   Den långa vägen- Den ekonomiska politikens begränsningar och möjligheter att föra Sveriger ur 1990- talets kris (The long road- The limits and possibilities of taking the Swedish economy out of the crises of the 1990s with Government policy

   1. T Andersson
   2. P Braunerhjelm
   3. B Carlsson
   4. G Eliasson
   5. S Fölster
   6. L Jagren
   7. E Kazamaki-Ottersten
   8. K-R Sjöholm

   (1993)

   Den långa vägen- Den ekonomiska politikens begränsningar och möjligheter att föra Sveriger ur 1990- talets kris (The long road- The limits and possibilities of taking the Swedish economy out of the crises of the 1990s with Government policy, Stockholm:IUI.

   • Google Scholar
4. 4
   Learning through short-run macroeconomic forecasts in a micro-to-macro model

   1. M Antonov
   2. G Trofimov

   (1991)

   IUI Working paper nr. 310. Published 1993 in Journal of Economic Behavior & Organization , 21:181–203.

   https://doi.org/10.1016/0167-2681(93)90047-S

   • Google Scholar
5. 5
   The Effects of Unemployment Compensation in General Equilibrium with Search Unemployment

   1. B Axell
   2. H Lang

   (1990)

   The Scandinavian Journal of Economics 92:531.

   https://doi.org/10.2307/3440390

   • Google Scholar
6. 6
   MOSES Handbook, Research Report No 35

   1. F Bergholm

   (1989)

   Stockholm:IUI.

   • Google Scholar
7. 7
   The Dynamics of Market Economies, North Holland, Elsevier Science

   1. R Day
   2. G Eliasson

   (1986)

   The Dynamics of Market Economies, North Holland, Elsevier Science.

   • Google Scholar
8. 8
   Business Economic Planning - Theory, Practice and Comparison

   1. G Eliasson

   (1976a)

   London, etc: John Wiley & Sons.

   • Google Scholar
9. 9
   A Micro-to-Macro Model of the Swedish Economy

   1. G Eliasson

   (1976b)

   Economic Research Report B15.

   • Google Scholar
10. 10
    Competition and Market Processes in a Simulation Model of the Swedish Economy

    1. G Eliasson

    (1977)

    American Economic Review 67:277–281.

    • Google Scholar
11. 11
    Micro-to-Macro Model of the Swedish Economy

    1. G Eliasson

    (1978a)

    Stockholm: IUI: Conference Reports.

    • Google Scholar
12. 12
    The Importance of Technology and the Permanence of Structure in Industrial Growth

    1. G Eliasson

    (1978)

    Relative Price Change and Industrial Structure- The “Norwegian Case, The Importance of Technology and the Permanence of Structure in Industrial Growth, IUI Conference Reports.

    • Google Scholar
13. 13
    Micro to Macro Model of the Swedish Economy

    1. G Eliasson

    (1978)

    How des Inflation Affect Growth? – Experiments on the Swedish Micro to Macro Model, Micro to Macro Model of the Swedish Economy, Stockholm: IUI, Conference Reports.

    • Google Scholar
14. 14
    Microeconomic Simulation Models for Public Policy Analysis, Vols. 1 &Amp; 2

    1. G Eliasson

    (1980)

    Experiments with Fiscal Policy Parameters on a Micro-to-Macro Model of the Swedish Economy, Microeconomic Simulation Models for Public Policy Analysis, Vols. 1 &Amp; 2, New York, London etc, Academic Press.

    • Google Scholar
15. 15
    Research Program and Proposal. Industriens Utredningsinstitut

    1. G Eliasson
    2. A Klevmarken

    (1981)

    Household Market and Non-market Activities, Research Program and Proposal. Industriens Utredningsinstitut.

    • Google Scholar
16. 16
    Policy Making in a Disorderly World Economy

    1. G Eliasson

    (1983)

    On the Optimal Rate of Structural Adjustment, Policy Making in a Disorderly World Economy, Stockholm: IUI, Conference Reports.

    • Google Scholar
17. 17
    Micro heterogeneity of firms and the stability of industrial growth

    1. G Eliasson

    (1984)

    Journal of Economic Behavior & Organization 5:249–274.

    https://doi.org/10.1016/0167-2681(84)90002-7

    • Google Scholar
18. 18
    The Firm and Financial Markets in the Swedish Micro-to-Macro Model - Theory, Model and Verification

    1. G Eliasson

    (1985)

    Stockholm: IUI.

    • Google Scholar
19. 19
    Technological Competition and Trade in the Experimentally Organized Economy

    1. G Eliasson

    (1987)

    Technological Competition and Trade in the Experimentally Organized Economy, Research Report No. 32, Stockholm: IUI.

    • Google Scholar
20. 20
    Evolutionary Economics Applications of Schumpeter’s Ideas

    1. G Eliasson

    (1988)

    151–199, Schumpeterian Innovation, Market Structure and the Stability of Industrial Development, Evolutionary Economics Applications of Schumpeter’s Ideas, Cambridge, New York etc, Cambridge University Press, p.

    • Google Scholar
21. 21
    The firm as a competent team

    1. G Eliasson

    (1990a)

    Journal of Economic Behavior & Organization 13:275–298.

    https://doi.org/10.1016/0167-2681(90)90002-U

    • Google Scholar
22. 22
    The Knowledge Based Information Economy

    1. G Eliasson

    (1990)

    “‘The Knowledge-Based Information Economy,’” Chapter I in Eliasson et al, 1990, The Knowledge Based Information Economy, Stockholm:IUI.

    • Google Scholar
23. 23
    Modeling the Experimentally Organized Economy - Complex Dynamics in an Empirical Micro-Macro Model of Endogenous Economic Growth

    1. G Eliasson

    (1991a)

    Journal of Economic Behavior and Organization 16:153–182.

    https://doi.org/10.1016/0167-2681(91)90047-2

    • Google Scholar
24. 24
    Deregulation, innovative entry and structural diversity as a source of stable and rapid economic growth

    1. G Eliasson

    (1991b)

    Journal of Evolutionary Economics 1:49–63.

    https://doi.org/10.1007/BF01202338

    • Google Scholar
25. 25
    Economic Growth and the Dynamics of Wage Determination - A Micro Simulation Study of the Stability Consequences of Deficient Variation in Factor Prices and Micro Structures

    1. G Eliasson
    2. T Lindberg

    (1986)

    IUI, Stockholm: Working Paper No. 170.

    • Google Scholar
26. 26
    A Micro to Macro Model of the Swedish Economy

    1. G Eliasson
    2. G Olavi

    (1978)

    Stepwise parameter estimation on a micro simulation model, A Micro to Macro Model of the Swedish Economy, Stockholm: IUI, IUI Conference Reports.

    • Google Scholar
27. 27
    A Micro-to-Macro Model of the Swedish Economy

    1. A Klevmarken

    (1978)

    On estimation and other problems of statistical inference in the micro simulation approach, A Micro-to-Macro Model of the Swedish Economy, Stockholm, IUI Conference Reports.

    • Google Scholar
28. 28
    The Markets for Innovation, Ownership and Control

    1. T Lindh

    (1993)

    “‘Lessons from Learning about Rational Expectations;’” forthcoming in Day, The Markets for Innovation, Ownership and Control, Amsterdam: North-Holland, Stockholm, IUI.

    • Google Scholar
29. 29
    The Phillips Curve and Labour Markets

    1. RE Lucas

    (1976)

    19–46, Econometric policy evaluation: A critique, The Phillips Curve and Labour Markets, Carnegie-Rochester Conference Series on Public Policy, p.

    • Google Scholar
30. 30
    Theories of Decision‑Making in Economics and Behavioral Science

    1. HA Simon

    (1959)

    American Economic Review 49:253.

    • Google Scholar
31. 31
    MOSES on PC, Manual, Initialization and Calibration

    1. E Taymaz

    (1991)

    Stockholm: IUI.

    • Google Scholar

Author details

1. Gunnar Eliasson

   Industrial Institute for Economic and Social Research (IUI), Stockholm, Sweden

   For correspondence

   g.eliasson@outlook.com

   "This ORCID iD identifies the author of this article:" 0000-0002-7967-6120

2. Erol Taymaz

   Middle East Technical University, Ankara, Turkey

   "This ORCID iD identifies the author of this article:" 0000-0001-7525-6674