The Nobel Laureate and the Learning Curve

The German scholar Clas-Otto Wene (Emeritus Professor at Chalmers University, Sweden) has built a worldwide reputation in recent years as an authoritative voice on the use of learning curves in analysis of energy systems. Readers of the Global Green Shift blog will recognize that learning curves are considered to be the foundation of future projections of cost falls in manufactured energy systems, and having particular relevance for the energy security of new industrial giants like China and India.

Clas-Otto Wene’s recent papers can be found at

(Kybernetes 2015) https://www.emeraldinsight.com/doi/abs/10.1108/K-01-2015-0014,

(Wiley Interdisciplinary Reviews. Energy Environment 2016)

https://onlinelibrary.wiley.com/doi/abs/10.1002/wene.172 and

(Futures 2018) https://www.sciencedirect.com/science/article/pii/S0016328717302008.

Most recently he has commented on remarks by Prof William Nordhaus, recipient in October this year of the Bank of Stockholm Prize in Economics (widely associated with the Nobel prizes). The official press release can be found at: https://www.nobelprize.org/prizes/economic-sciences/2018/press-release/

Professor emeritus Otto’s Note was circulated privately to a small group of scholars. With his permission, the Note is reproduced here as part of the GGS blog.

The Laureate awarded the 2018 Prize in Economic Sciences in Memory of Alfred Nobel, William D. Nordhaus, states that learning curves “have a weak empirical foundation and can lead to misleading results for policy” (Nordhaus, 2014). The reason is that there is “a fundamental statistical identification problem in trying to separate learning from exogenous technological change”.

Those are strong words. If true, they raise doubts on the legitimacy of Governments´ interventions in technology markets to foster deployment of low-carbon technologies. If true, the transition away from fossil technologies may be very costly, indeed. This means that large damage costs must be tolerated, corresponding to considerable rise of earth average temperature.

I agree with Nordhaus about the futility in using learning curves to distinguish effects of endogenous technology learning and exogenous technological change. But Nordhaus conclusion does not follow. From the cybernetic perspective, the learning curve has a strong empirical foundation and manifests the efficient working of an autonomous learning system (Wene, 2015).[1] I demonstrate this point by using public R&D as an example of a vector for exogenous technological change (Wene, 2016, pp.24-26).

Public and private R&D have quite different identities; the former seeks to enhance the capacity for effective action in society, the latter in the enterprise or in the enterprises to which it belongs. Private R&D is therefore part of the network of internal operations within the system while public R&D is an external process found in the system environment. Public R&D cannot directly influence cost but needs being taken up in the private R&D loop for its results to become part of the industry knowledge stock (Watanabe et al., 2000). The industry learning system has operational closure (Varela, 1979, 1984), meaning that it is an autonomous unit in full control of all its operations, including those of its own R&D. The learning curve measures the effect on system performance from these interconnected operations. The operations process and absorb relevant public R&D findings, which therefore must not be considered separately. Overwhelming empirical evidence supports this interpretation of the learning curve. The evidence shows learning curve pervading all levels of industrial activities, e.g., the same learning curve steers building of man-of-war for the Swedish navy in 1780s, global production of PV-modules in the latest five decades and decarbonisation of the world economy over the last 150 years.[2]

The operationally closed learning system handles any other vector for exogenous technology change in the same way as public R&D. Operational closure, however, creates a paradox. It moves the focus away from the object to the acting subject, from the produced technology to the learning system and its organisation. But the learning curve tells us that the learning system is a self-organising system that seems to successively increase order, meaning that it continuously decreases entropy in obvious contempt of the Second Law of Thermodynamics. In his seminal paper from 1960, von Förster (1960) therefore formulates the paradox: “There are no such things as self-organizing systems!” The resolution of this paradox comes from the path-breaking work in non-equilibrium thermodynamics by Onsager (1931) and Prigogine (1947, 1980), Nobel Laureates in Chemistry in 1968 and 1977, respectively. Applying their findings provides the shape of the learning curve and learning rates in agreement with observations (Wene 2013, 2018). In equilibrium markets the learning curve traces the path of minimum entropy production, i.e., provides optimal use of a scarce resource, and has a basic learning rate of 20%. The pervasive learning curve thus emerges as a reliable tool with a strong theoretical foundation in modern thermodynamics. The conclusion based on the work of the two Chemistry Laureates is therefore quite different from that of the 2018 Laureate in Economic Sciences.

A modelling experiment in mid-1990s at Chalmers University of Technology sheds some light on the effect of learning curves on the cost of a transition away from fossil fuels (Mattson and Wene, 1997). The total costs for the global electric system over a 50-year period were the same for a system based on advanced coal technologies (“clean coal”) and a system where coal was phased out by deployment of solar photovoltaic, electricity storage and highly efficient natural gas technologies.[3] The results were discussed in a publication from the International Energy Agency (IEA, 2000, pp. 84-91) and were at that time met by some internal scepticism. However, today this modelling experiment should be criticised not because of lack of realism but because of lack of ambitious goals. The catch here is that the ride down the learning curve to reach competitive cost for new technologies needs considerable learning investments up front. To start this ride therefore requires trust in a stable learning curve. But through audacious government deployment programmes in, e.g., Japan, USA, Germany and China, PV-technology has reached low-cost much faster than envisioned in the modelling experiment.

Two conclusions can be drawn from the discussion above. Firstly, although costs are important for short run decisions, in the long perspective they are a function of political will. This means that the concept “optimal temperature rise” has no meaning outside the domain of political decisions. However, all decisions happen in a Westphalian World where all participants set their national interests first. This brings us to the second conclusion. Secondly, we find from experience with solar PV and wind that the learning curve works very well in a global Westphalian World while agreements on emission caps or taxes do not. The reason is that exploiting the learning curve makes nations compete for technology leadership. Focusing on policies to foster technology change will therefore have a better chance of bringing a successful transition away from fossil fuels than policies focused on caps and taxes.

Nordhaus critique of the learning curve illustrates the ambiguous relation of neoclassical economists to the pervasive endogenous technology learning measured by the curve. Arrow (1962) introduced the curve into neoclassical economic theory in a pioneering paper, as noted by the Committee that awarded him the Economy Prize in Memory of Alfred Nobel in 1972. However, he restricted his analysis to the Labour production factor although he foresaw also curves for other production factors. It was the Boston Consulting Group (BCG, 1968) which widened the scope to curves for total costs and thus opened for the idea of the learning curve pervading all levels in industry. Accepting the pervasive learning curve as an agent of technology development may, however, cause some collateral damage to the to the neoclassical theory building. The pervasive learning curve indicates that the set of future possible technology solutions is truly non-convex. However, convexity is a necessary condition for applying the so called Second Theorem of Welfare Economy, which forms the theoretical base for deregulation of former state monopoly markets, e.g., electricity production. This raises doubts about the long-range efficiency of deregulated markets.

References

Arrow, K. (1962). “The Economic Implications of Learning by Doing”, Review of Economic Studies, p. 155.

BCG (1968). Perspectives on experience. Boston, Mass: Boston Consulting Group.

IEA (2000). Experience curves for energy technology policy. Paris: International Energy Agency/Organisation for Economic Co-operation and Development.

Mattson, N. (1997), Internalizing technology development in energy system models, Thesis for the Degree of Licentiate of Engineering, Energy Systems Technology, Chalmers University of Technology, Göteborg, Sweden.

Mattson, N. and Wene, C-O. (1997). “Assessing New Energy Technologies Using an Energy System Model with Endogenized Experience Curves”, Int. J. of Energy Research, Vol. 21, p. 385.

Nordhaus WD. (2014), “The perils of the learning model for modeling endogenous technological change”. Energy J 35:1–13.

Onsager, L. (1931), “Reciprocal Relations in Irreversible Processes. I”., Phys. Rev. 37, 405 - 426.

Prigogine, I. (1947), Etude Thermodynamique des Phénomènes Irréversibles, Dunod, Paris.

Prigogine, I. (1980), From being to becoming. Time and Complexity in the physical sciences, W.H. Freeman and Company, New York.

Varela, F. (1979), Principles of Biological Autonomy, Elsevier-North Holland, New York.

Varela, F. (1984), “Two Principles for Self-Organization”, in: H. Ulrich and J.B. Probst (eds.), Self-Organization and Management of Social Systems, p.25-32, Springer, Berlin.

von Förster, H. (1960), “On Self-Organizing Systems and Their Environment“, in: M.C. Yovits and S. Cameron (eds), Self-Organizing Systems, pp. 31-50, Pergamon Press, London. Reprinted in H. von Förster, Understanding Understanding, 2003, Springer, N.Y., Berlin, Heidelberg.

Watanabe C, Wakabayashi K, Miyazawa T. (2000), “Industrial dynamism and the creation of ‘virtuous cycle’ between R&D, market growth and price reduction – the case of photovoltaic power generation (PV) development in Japan”, Technovation, Vol. 20, pp. 299–312.

Wene, C.-O. (2013), “Learning Curves tracing the optimal path for Technology Learning Systems”, in: Proceedings of the 13th IAEE European Conference, Düsseldorf, Germany, 18-21 August, 2013, available from: http://www.iaee.org/en/publications/proceedingssearch.aspx

Wene, C.-O. (2015). “A cybernetic view on learning curves and energy policy”, Kybernetes, Vol. 44:6/7, pp. 852–865.

Wene, C.-O, (2016), “Future energy system development depends on past learning opportunities” WIREs Energy Environ, Vol 5:1, pp. 16–32, doi: 10.1002/wene.172.

Wene, C.-O. (2018), “Quantum Modelling of the Learning Curve”, Futures, Vol. 103, Pages 123-135

Yeh S, and Rubin E. (2012), “A review of uncertainties in technology experience curves”, Energy Economics, Vol. 34, pp.762–771.

[1] To avoid confusion: by “learning curve” I mean the classic single-factor learning curve. The hybrid two-factor or multifactor learning curves do not escape the critique of Nordhaus. For discussion of the hybrid learning curves, see Yeh and Rubin (2012).

[2] See presentation to the IEW 2018, https://iew2018.org/wp-content/uploads/2018/07/1G_Wene.pdf . IIASA finds a learning curve for decarbonisation of the US economy between 1850-1990 (see IEA, 2000, p. 76)

[3] No CO2 tax or cap were imposed on either system and the conditions for the two systems were identical. Real rate of interest is 5%. However, introducing learning curves makes the system strongly non-linear and many (“local”) optimal systems emerges. Technically, the model was solved by a perfect-foresight algorithm minimizing cost over 50 years. Private investors will not have such long time horizon, that is why government deployment programmes are needed to raise learning investments. For model details, see Mattson, 1997.