The New Economics of Industrial Policy2024

There has been a revival of industrial policy in economics. Previously there were negativa and vicious reactions from mainstream economists against this idea. Industrial policy has many forms but can always be defined as creating incentives for private-sector behavior to act in a manner that is consistent with some notion of structural change. There are three rationales given for industrial policies.

The first are positive externalities of learning, national security, and of good-jobs. These are best handled by Pigouvian subsidies. The second is the coordination market failure, where individual producer's profitability may depend on other producers down or upstream the linkage chain. It may be that there are suboptimal equilibriums when there is no coordination. This is best resolved by governments calling on firms to be together or make simultaneous investments. Finally, we have public inputs, in which governments are doomed to choose (insofar as having to make choices of building infrastructure, prioritizing x or y training, etc.)what activities are more deserving of these goods or inputs. These require the provision of the government, evidently.

At the side of these strong cases for industrial policy, here are also strong critiques. They usually center around political capture and information shortcomings. The first describes lobbying, corruption, and so on affecting the interventions in a way that distorts social profit into the benefit of these self-interested actors, and the second says that governments are not expected to know enough of the previous market failures to correctly act upon them. There are other more theoretical critiques, for example, if inelastic demand leads to modest aggregate welfare gains, despite strong external economies of scale, it can be said that industrial policy does not lead to structural change.

It is difficult to measure industrial policy. The first is that it requires many tools, that usually change over time. Some scholars conflate industrial policy with protectionist trade policy, and this is not really the case in contemporary settings. There have been efforts to distinguish government expenditures, the use of ML to classify policy descriptions, etc. This allows researchers to capture policy that does not require spending, examples of coordination policy, distinguish industrial policy from other policy, etc. These efforts show that industrial policy is on the rise, with higher-income countries being major users. In lower-income countries it may be that fiscal constraints bind fast for industrial policy that requires spending. Countries tend to do the same across the income distribution.

They then start with he empirical challenge in evaluating industrial policy. Let \(g(\theta)=(1-\theta)A\) be the government performance as a negative function of a parameter of market failure \(\theta \in (0,1)\), with \(A\) being some state variable affecting performance. Teh greater the market failure, the larger the gap between social and private returns. Let \(s \in [0,1]\) be a subsidy that comes with some fiscal cost \(\varphi \alpha (s)\) with \(\alpha\) being a rising and convex function of \(s\) and \(\varphi\) being the ability to intervene effectively. The subsidy alleviates the gap to \((1-\theta)[1-s]\). Then, with industrial policy we have growth being \[g(s,\theta,\varphi)=(1-\theta(1-s))A-\varphi\alpha(s)\] The optimal level of the subsidy \(s^{\text{soc}}\) is the value of \(s\) that solves \(g_{s}(s,\theta,\varphi)=\theta A - \varphi^{\prime} \alpha(s) = 0\). However, there are other motivations for the subsidy. The government can choose to channel resources to favored groups, Salinas de Gortari style. Let \(\pi(s)\) be a single-peaked, concave function, with max value \(\pi(s^{\text{pol}})\), that represents political benefits to the government of using the subsidy. Assume government behavior maximizes a value function \[u(s;\theta,\varphi)=\lambda g(s,\theta,\varphi)+\pi(s)\] where political and economic objectives enter, the latter with weight \(\lambda\). The government chooses the level of subsidy \(s^{\text{gov}}\) which is the solution to the FOC \[\lambda g_{s}(s,\theta,\varphi)+\pi^{\prime}(s) = \lambda [\theta A - \varphi \alpha^{\prime}(s)]\pi^{\prime}(s)=0\] We can now consider how observational data can help us see if industrial policy works. It would ideally help us distinguis three views: governments can succesfully identify and support firms that will enhance whatever objective (developmentalist), governments try to support firms but do so inappropriately (inefficacy), and governments do not seek desirable outcomes and are basically corrupt (rent-seeking). We only observe how \(g\) varies with \(s\). The first generation of empirical work was mostly correlational, and infered industrial policy was mostly ineffective. However, with this model, the lower the \(\lambda\), the larger the subsidy and the lower the rate of growth. This confirms rent-seeking. Howeverm if the only determinant of the policy was market falures \(\theta\), then governemnt is a social-welfare maximizer. In this case, internention is higher with higher market failure, and the larger the market failure the lower the growth rates, as socially optimal policy does not totally offset the market failures. If you notice, this is the same pattern as the rent-seeking government, despite this particular government being the opposite. Thus, observational data can be misleading.

The modern empirical toolkit is better suited at answering and evaluating the efficacity of implementation. There is new evidence of this kind on sectoral industrial policy, maybe one of the most controversial types. There is a new crop of studies, in particular Lane's paper on HCI drive, and Juhász's article on the Napoleonic blockade, that seem to support the infant industry hypothesis. It could be that temporary protection changed the long-term profitability of production across regions. There is also some work that finds that tacit knowledge was an important component of technology transfer. They believe the literature must move towards questions of implementation, as studies are too abstract to inform policy in any way. Even Lane's QJE paper cannot distinguish from the HCI bundle what was it that led to increased comparative advantage in South Korea. There is urgent need of more work informing what makes for effective sectoral industrial policy. Public R&D may have a place in the toolkit, and that place-based industrial policy can shift the composition of local economic outcomes. There is also some work on Liu (2019) using a network model to show that the optimal sectors to target are those whose imperfections are the most consequential given their impact through input-output linkage. This is all ina section analysing the East Asian miracle. There is then some welcomed sociology from Peter Evans, who argues fro embedded autonomy, which is that close links between private and public sectors may not always lead to capture, but might be necessary in ensuring proper information to design policy. There is some more on Bartik and training being more effective than tax credits, and on the fact that there is no reason to be solely concerned with manufacturing when it comes to industrial policy.

Overall a great article, easy to read, and very relevant to the work I am currently doing.