Working Paper

Evaluating the Learning-by-Doing Theory of Long-Run Oil, Gas, and Coal Economics

May 23, 2017 | Justin Ritchie, Hadi Dowlatabadi


Energy production techniques have undergone extensive technological change. To understand future dynamics, long-term studies adapt a learning-by-doing model from manufacturing to understand productivity gains. We examine the suitability of this approach.

Key Findings

  • Long-run fossil energy supply modeling can reflect the economics of how firms in manufacturing industries learn over time and how that learning affects cost projections—but may overestimate productivity gains.
  • We examine this approach to assess whether a non-price–induced learning effect influences international oil and gas productivity trends since the late 1970s, using data from the US Energy Information Administration’s Financial Reporting System.
  • We find that productivity gains occur independent of the market price in oil and gas operating expenditures through the early twenty-first century, confirming the usefulness of a learning model for this aspect of industry cost.
  • But capital expenditures begin to dominate industry spending after the mid-1990s (diverging from what we expect from a learning model), likely due to the influence of price-sensitive development expenditures that prepare reserves for production.
  • Overall, we find a compounding learning effect applied over many decades to assess economically accessible reserves is likely to overestimate their total net benefits and underestimate equilibrium market prices.
  • Nordhaus (2009) developed a critique of the learning-by-doing model for energy technologies that anticipates this outcome. We adapt that work to examine the difficulty of using learning effects to understand productivity in fossil energy resources.


Energy and climate policy studies with a long-term outlook need to anticipate potential developments in technology and the temporal nature of today’s resource-reserve definitions for oil, gas and coal. Accordingly, economic concepts of learning formulated from research on manufacturing industries inspire a common approach to modeling technological change in hydrocarbon energy resource production. This theory expects future costs of fossil energy supply to benefit from a cumulative learning effect which results from ongoing extraction. With three decades of data since the initial formulation of this theory by Rogner (1997), some key regions of conventional oil and gas production have matured. Fresh data on industry cost trends are now available, allowing for a closer examination and validation of whether this learning model hypothesis is relevant for long-run cost projections. Empirical cost and productivity data challenge the broad application of a learning model to the total geologic occurrences of fossil energy resources. We find that oil and gas industry operating costs indicate a learning effect, but capital expenditures do not. Coal resource-reserve dynamics have not developed as anticipated. Nordhaus (2009) suggests technological change models of energy supply calculated with a learning curve will consistently overestimate productivity gains, producing biased cost estimates of future technologies. This paper considers the Rogner (1997) learning-by-extracting model for fossil energy supply as a specific case of Nordhaus’ argument.