How to Choose Among Technologies With Learning Curves: Making Better Investment Decisions

Abstract

Learning curves, the fact that technologies improve as a function of cumulative experience or investment, are desirable-think inexpensive solar panels or higher performing semiconductors. But, for firms that need to pick one technology among several candidates, such as R&D labs or venture capital firms, learning curves make it hard to make the optimal technology investment choice. This paper addresses this challenge and provides the first formal analysis of how to invest in a set of technologies over multiple periods if (i) the cost of a technology decreases with cumulative investment and (ii) the decision maker only benefits from the estimated lowest-cost technology at the end of the investment horizon. We develop a dynamic programming framework to model this problem and, for a 2-technology case, are able to identify a closed-form ratio that managers can compute to choose the optimal technology to invest in. We then leverage these insights to develop a policy that can be employed to find the best investment choice for settings with any number of technologies and decision periods. Our policy has the added benefit of making the value of exploration versus exploitation visible to the decision-maker. We provide detailed numerical comparisons of our Technology Learning Curve Optimization (TELCO) policy and find it performs, on average, 8% better than the second-best policy across 1200 different scenarios. Additionally, we apply our methodology to the real-world problem of lithium-ion chemistries for battery production, demonstrating its robustness and effectiveness.

Keywords: learning curves, technology, innovation, batteries, energy storage, sequential decision making, TELCO, exploration, exploitation