Skill transferability and the stability of transition pathways- A learning-based explanation for patterns of diffusion

Abstract

Understanding and governing technology transitions is essential to cope with major challenges of the 21st century such as climate change or digitization. In this paper, a learning-based approach is developed to explain the dynamics of different transition pathways. Technological know-how is necessary to make effective use of technical innovations embodied in capital. Firms and employees accumulate technology specific knowledge when working with specific machinery. Radical innovation differs by technology type and pre-existing knowledge may be imperfectly transferable across types. This paper addresses the implications of cross-technology transferability of skills for firm-level technology adoption and its consequences for the direction of macro-level technological change. A microeconomically founded model of technological learning is introduced. The model is based on empirical and theoretical insights from the innovation literature. In a simulation study using the macro-economic ABM Eurace@unibi-eco and applied to the context of green 2 technology diffusion, it is shown that a high transferability of knowledge has ambiguous effects. It accelerates the diffusion process initially but comes at the cost of long-term technological stability and specialization. For firms, it is easy to adopt new technology, but also easy to switch back to the incumbent type. Technological instability can be macroeconomically costly.

Appendix A: Simulation results

1.1 A.1 Settings

Table 3 Simulation settings

1.2 A.2 Baseline

Here, only some general features of the simulated time series data are shown. Information about the empirical validation is available in SM.I. More detail on this baseline simulations is available in Hötte (2019b). A longer discussion of a similar simulation is provided in Hötte (2020a). A difference to the simulations in Hötte (2020a) is given by lower diffusion barriers of the green technology and another specification of the learning function. The simulation model, the simulated data and a selection of results of descriptive statistics is available in a separate data publication (Hötte 2019d).

In Fig. 7, the time series are dis-aggregated into green, conventional and so-called switching regimes. A simulation run is classified as switching regime if the diffusion process is very volatile, i.e. when ${\nu ^{c}_{t}}$ heavily fluctuates between low and high levels or when ${\nu ^{c}_{t}}$ has not converged to more than 90% or less than 10% in T (cf. Hötte 2020a). This is associated with uncertainty about the final technological state. The time series data illustrate that “technological uncertainty” is costly in terms of aggregate (log) output (Fig. 7j). It is associated with wasted resources because R&D and learning time are invested in a technology that becomes obsolete in the long run. This leads to a delayed technological specialization compared to the green or conventional regimes with a more clear-cut technological path selection. Figure 7d and e show that this is associated with a delayed divergence in relative knowledge stocks which is a reason and a result of uncertainty.

Fig. 7

Time series of macroeconomic and technological indicators. Different line shapes indicate different regime types (: transition, : conv, : switch)

Figure 7g shows the evolution of relative prices for capital goods and Fig. 7h shows this price normalized by the relative productivity. Figure 7f illustrates the price for material resource inputs normalized by real wages. The price evolves such that it accounts for roughly 9.5% of wage costs for an average firm during the whole time horizon.

Figure 7c illustrates an alternative environmental performance measure, called eco-efficiency, that measures the environmental impact (here amount of natural resource inputs) per unit of produced output. The eco-efficiency also improves when the productivity performance in the lock-in regime improved by technical progress which is a relative decoupling of production from environmental damages. Principally, there can be a trade-off between the specialization in the conventional technology and the switch to green technology if the success of the transition is uncertain. However, modeling this trade-off is very sensitive to the modeling assumptions regarding the environmental impact, initial conditions about the available technology options and productivity improvements in both sectors. In this study, the focus is on replacement dynamics in a theoretical way which allows to be agnostic about assumptions that may critically affect such a trade-off analysis.

A two-sided Wilcoxon test indicates that the differences between green and conventional regimes a (cf. Fig. 7) are significant (see SM.I.3).

1.3 A.3 The ease of learning

The pace of relative technological learning is also dependent on the technological difficulty χ^int. If a technology is very easy to learn, i.e. χ^int = 0, the learning progress is independent of the time invested in learning which is proxied by $\nu ^{c}_{i,t}$. If χ^int is high, the progress is sensitive to $\nu ^{c}_{i,t}$, i.e. learning is more effective if employees work only with one technology type. In an experiment that is longer discussed in Hötte (2019c), it had been shown that χ^int is only of minor importance in the presence of cross-technology spillovers.

The impact of the difficulty on the learning speed is most critical in times when firms are transitioning to alternative technology. During a phase of technology change, a trade-off in the allocation of the learning time exists. This trade-off is more pronounced when a technology is difficult to learn. A technology that is easier to learn is associated with lower technology switching costs. This may have an ambiguous effect on green technology diffusion. It is easier to switch to green technology, but it is also easier to switch back if the difficulty is symmetric. Whether increasing returns to learning stabilize an ongoing diffusion process, depends on the extent to which the green technology is adopted in the first years.

The adoption in the early phase is facilitated by cross-technology spillovers reflected in a lower distance χ^dist. If the transferability is sufficiently low, increasing returns to learning contribute to the stabilization of the technological regimes.

1.4 A.4 Interactions between spillovers and the ease of learning

In the Monte-Carlo experiment in Section 4.3, the learning parameters are drawn at random, i.e. χ^dist ∈ [0, 1] and χ^int ∈ [0, 2]. Diffusion barriers at the day of market entry are fixed at a level of 3% (β^A = β^b = .03) as before. The transition probability accounts for 64%.

In Table 4, means and standard deviation of the initialization are summarized as aggregate and dis-aggregated by regime subsets. The p-value in the last column indicates whether the difference in means between the two regime types is significant. The average mean of the distance χ^dist is significantly lower in the subset of green regimes. The difference in the χ^int is only weakly significant at a 10% level. Some general descriptive information of these simulations is provided in SM.III.2.

Table 4 Initialization of learning parameters