13. Theories, Choice of Models, and Estimation Methods: The Problem of Empirical Credibility

See further, for instance, Eliasson (1978b, 1983a, b, 1984, 1991b).

The phenomenon of deterministic chaos was recognized in the 1970s by meteorologists, who experimented with nonlinear empirical models and understood that the generated “surprise behavior” could be what they recognized from their weather charts. The phenomenon as such was however discovered as a mathematical property more than 100 years ago by the French mathematician Henri Poincare (1891) who studied the so-called three-body problem (Eliasson 1983a, b, Puu 1989). The same phenomenon obtains in an economic model that cannot be solved for a price/quantity equilibrium, which becomes the situation if many interacting agents, as in the micro to macro model (of Sect. 13.6), compete to improve their profit positions, and the structure of the economy constantly changes because of that competition. This is the case if the population of agents changes endogenously through endogenous entry and exit (selection). Such selection models are often highly nonlinear. Sometimes competition is so intense that the economic system may come close to or hit an infinite regress and collapse (Eliasson 1984, 1991b).

The quality and credibility of managing an economy through policy is normally judged by the same criteria as those applied in medicine. If conducted according to “scientific knowledge and accepted experience,” the medical treatment is considered to have been responsibly conducted. Failed treatments cannot then be legally taken to court for misconduct, even if the treatment is later found to have been entirely wrong. So we have a problem in the policy-making field, if received scientific method and exercised policy practice are based on a model the a priori specification of which can be demonstrated to be empirically without a credible foundation. Who is responsible if policy makers wreck the economy?

To be observed is also that a radical turnaround of Swedish policies in a significantly more entrepreneurial direction (as argued to be necessary in Andersson et al. 1993) was enacted from around 1993 and seems to have been sufficient for Swedish manufacturing to catch up on lost ground in a few years. This can also be seen in Fig. 1.​1c.

If so, they are not (by definition) evolutionary (see Chap. 11). Equilibrium trajectories can however be imposed, and then we are back in the conventional neoclassical camp.

Investment decisions are micro based in the markets for manufacturing products. Private consumption is fed back into all markets when endogenous incomes are filtered from firms through an aggregate Stone-type expenditure system, which has been “upgraded” by nonlinear, household savings, and investment features. See Sect. 13.6 for further references. Plans to integrate a micro household system in the micro to macro model (Eliasson & Klevmarken 1981) have so far not been realized, even though the model was prepared for that from the beginning

The latter was the case when micro to macro modeling work began in 1974. We carved out four manufacturing sector aggregates in an existing 11-sector Keynesian and Leontief model, reorganized the entire micro to macro data flows onto a market, rather than production defined format (to correspond to the taxonomies that firms used in their own statistical systems, that we tapped into through the special surveys conducted annually), and replaced the four market aggregates with real firm data (Eliasson 1976b, 1977, 1978a, 2014d)

The original Romer (1986) story is different in that Romer was mainly concerned with overcoming the static diminishing return hurdle of the neoclassical model. In the short run, he allows for strictly diminishing returns to R&D investments, but in the long run everybody learns from one another, thus shifting the productivity of the entire production system upward, a micro to macro effect that is not explicit, but represented by aggregate coefficients in his macro model.

Or rather Eliasson (2010a)

Simply because competitiveness is so much more than technology. Once invincible firms always eventually get sloppy and lose their competitive edge or make some disastrous mistaken choice. I often use IBMs supreme performance during the 20-year period 1965 to the mid-1980s, when its “supremacy” abruptly ended (Eliasson 1996a: 179 ff) as an example. The interesting thing is that IBM management knew quite well in principle what had to be done, but that its dominant “mainframe computing” mind was uncapable of achieving the needed change in mindset in practice. IBM management finally managed a turnaround by importing outside top level (non-computing) management, and reduced total staff to about half. IBM is still one of the major IT firms in the world but no longer a technological leader. The story of Kodak did not have the same happy ending. Kodak, the world leader in chemical photography, could not transform itself from a chemical to an IT company. It understood that it had to, and tried, but failed and has been closed down.

This is a common ceteris paribus assumption/trick to keep prices exogenous throughout the analysis and to eliminate the distinction between micro and macro. In a properly specified dynamic model, it should not be possible to hold prices and structures unchanged during a technology change of some order of magnitude. However, allowing for simultaneous change of prices and quantities takes you outside the mathematics routinely used by economists. The firm is simply assumed to be so small that whatever it does, market prices will not be affected. Hence there will be no incentive to reallocate resources. Aggregation becomes static, and the dynamic problem addressed is turned into an accounting exercise. Reallocation of resources, however, is exactly what occurs when technologies change and what we should be concerned about. Such dynamics is what can be handled analytically using simulation mathematics. To understand the macroeconomic consequences (that may be very large in the long run) of even a small reallocation of resources, selection, and structural change, nonlinear modeling is needed.

In case the reader wonders what an infinitesimally small or atomistic firm might look like, just think in terms of (13.6), and imagine the scale to be reduced infinitely. Even though the firm vanishes out of sight, it may still be represented by the infinitely small inputs (A, X) generating infinitely small output in (13.6).

The analogy with accounting is again illustrative. The accounts measure the final outcome of a dynamic business process as it looks at one point in time, but do not contain the information needed to explain how the accounts came about.

Note that this was also assumed in the original innovation systems “models” of Freeman (1974, 1982), Lundvall (1992), and Nelson (1992a, b) in which growth was directly R&D driven. See further next Sect. 13.4.

Please observe that this is the exact opposite to the conclusion of neoclassical general equilibrium theory, in which the zero transaction costs assumption makes centralized cost minimizing optimal policy superior. There is then no account of lost winners due to imperfect or incompetent selections, because they cannot occur by assumption. Under the assumptions of the Experimentally Organized Economy (EOE), a fairly complete competence bloc is needed for the allocation mechanisms to at all function, and with significant transactions costs explicit, an entirely different economic world emerges (Chap. 11).

Even though economists tend to see the world Y explained by Y = F(x) + ε, where x is a vector of explanatory variables, and ε is a random variable with nice stochastic properties, that picks up the influence on Y of all that the X vector fails to “explain.”

A quarter was regarded as a “normal” production planning period when modeling work was initiated in 1974, and the empirical implementation of the model is based on that period specification. Today quarterly production planning in manufacturing firms, increasingly using computer-based continuous planning may sound a bit outdated. But period definition is an arbitrary choice and can easily be technically changed to a shorter period setting and also empirically if sufficient data are available

The micro to macro model was in fact used at CEMI in Moscow at an early stage of development on a highly simplified form (to be compatible with available computer capacity on a Norsk Data mini-computer) to simulate the transition from a centrally planned to a market economy. See Altaev et al. 1982, 1987, 1989. A related and more advanced simulation analysis on the Swedish version of the model was later published by Antonov and Trofinov (1993)

Quite in line with Stockholm School economics the positions and ex ante plans of firms that make up the initial state at one point in time are never consistent in the sense of being in a market-clearing static equilibrium guaranteeing that ex post outcomes are identical to ex ante plans. The ex ante inconsistencies are therefore a critical element in micro to macro economy wide dynamics. It is also important that these inconsistencies are not confused with measurement errors in the database. Both create “dynamics” in micro to macro model simulations.

Since I was at the time the head of that department and initiator of the survey, I could also design the questions to firms to be compatible with both their internal statistical systems, as studied in Eliasson (1976a), and the requirements of the micro to macro model

Or for that matter an accounting general equilibrium model of the type developed by Tobin (1969). Also see Bezemer (2010) on these kinds of models, and Eliasson (1985) on how a stock-flow consistent Tobin type of flow and asset accounting scheme has been integrated in the Swedish micro to macro model.

In practice enforced as an “exogenous equilibrium” that is not affected by the policy experiments conducted on the model, very much as similar long-run rational expectations equilibria are enforced in DSGE models (See Sect. 13.6.12).

See Sect. 9.​8.​1.​

Several companies in the database are composed of several divisions, each of them currently represented in the model as an individual firm. The idea was to implement the business financial system in Fig.13.2 as a corporate headquarter (CHQ) financial management and long-term investment and coordination management module (Eliasson 1985;Supplement II:15ff), which so far has not been done.

That survey was headed at the time by Bo Carlsson and reported on in Carlsson et al. (1979).

We redefined and reorganized the Swedish national accounts (NA) on a final goods format to correspond to determinate article OECD end-use classification code (see Ahlstrom 1978, Nordström 1992, and in general Albrecht et al. 1992). The main reason was to obtain a market classification of NA data that was relevant for business decisions and above all compatible with the internal statistical systems of firms that we accessed through the planning survey of the Federation of Swedish Industries to load the micro to macro model with real firm data

The USAGE model is the USA derivate of the Australian MONASH CGE model.

Again, my comments are based on what Rand (2015a: pages x and 10) reports. I had to guess because the Australian Department of Defense that commissioned the Deloitte (2014) study has not been willing to let me see it. See further Sect. 9.​8.

See discussion introducing Sect. 10.​6.

Or the state space of the corresponding mathematical model

The algorithms that govern market processes in the micro to macro model push firms to increase their returns to capital or their market valued net worth. If they succeed they raise their ambitions. If they fail they gradually lower their ambitions. Variations of this maintain or improve profits (MIP) targeting principle were universally applied in the budgeting practices of large firms interviewed in Eliasson (1976a). If returns came down too much in the model (the interest rate defines a bottom profitability target), firms have to act. If they keep failing, their losses eat into their net worth, and they are eventually forced to exit. If there is no counteracting mechanism in the form of entry, the firm profitability distributions (the Salter curves in Fig. 11.​1b) will be flattening out, and when all є = R-r (in 2B) have come too close to zero, or financial market equilibrium, the whole model economy becomes destabilized, because such a market-clearing solution does not exist as an operating domain of the model (Eliasson 1991b). Profit maximization is a limiting special case of the MIP principle. See further Eliasson 2014d.

There is an interesting “parallel” to this. Kenneth Arrow, in his well-known learning-by-doing article (1962a) was inspired by Erik Lundberg’s (1961) study of the Swedish Horndal steel works, in which a steady increase in productivity had been registered over the years, even though no investments had been made during the same period. Lars Vinell (1998) went to a neighboring steel works (the Horndal works had been shut down and the archives lost), which had had a similar experience. He found that the reason for the steady increase in productivity was that management had changed the composition of qualities of the aggregate steel product of the plant toward those that captured the highest prices and even shut down the production of some steel qualities. When the aggregate value so obtained was deflated by the official index for quality steel, the Horndal effect of Erik Lundberg appeared.

Applying the MIP profit targeting principle explained in Sect. 13.6.2.

Notably for the initial year of a simulation run. Albrecht et al. (1992) present the measurement principles behind that database and a full account of the initial year 1982.

The way we imposed that global discipline of a static equilibrium on the micro to macro model is similar to the equilibrium method Kydland and Prescot (1982) used to calibrate their macro model, only that the markets of the micro to macro model were never in static or rational expectations equilibrium.

Experiments with endogenous entry and exit will be shown in Sect. 13.10 below. Also see Eliasson (1984).

The documentation, unfortunately, does not include simulated output for the entire national economy, which would have made it possible to say something on the Keynesian demand effects under those conditions. While unemployment is somewhat higher initially, when government reduces its hiring rate, there is no difference toward the end of the simulation.

For instance aircraft safety engineers in Saab to Autoliv (Eliasson 2010a:151f). Investments in aircraft safety embodied as human capital in these engineers helped develop new innovative automotive safety products in Autoliv. Medical spillovers from weapons development are another example (Chap. 5). It was rarely a case of an identifiable technological spillover from Saab to a medical instrument company, but of engineers who had learned on a military project who had understood the medical potential, and, for instance, started a new company.

Just for the record. Among all the telecommunications companies attempting to enter the digital mobile telecom market in the 1980s, most failed. Only two were successful (Ericsson and Nokia), and both had an in-house military radiotechnology (Eliasson 1996a).

Reveals former Ericsson CEO Carl-Henric Svanberg in a new book Ericsson’s Kris och Resan Tillbaka (2015) written by Ericsson’s former press secretary Åse Lindskog. Also see Ny Teknik , 28 Oktober 2015.Nr 44:14–15.