The *ex ante* approach, of course, has its own limitations. However, it can map out the frontier of policy outcomes. Strikingly, in many cases, even optimally designed policies yield only modest gains, which is informative in itself.

link to article:
alashkar.pages.iu.edu/New_Industri...

13. We then clarify the logic behind *ex ante* policy evaluation, a growing yet misunderstood method.

At its core, this approach involves estimating elasticity parameters using in-sample price & quantity data, using them to simulate out-of-sample quantities under optimal prices

(b) Heterogeneous treatment effects

Industrial policy is justified because the marginal return to adding resources (VMPL) differs across units. Otherwise, there's no rationale for policy. Hence, contexts where industrial policy is warranted are inevitably characterized by HTE.

11. The identification challenges are twofold:

(a) SUTVA violation
By design, industrial policy reallocates factors to treated units by drawing them from non-treated units. So, spillover effects are inherent to the policy itself.

10. Design-based evaluations face two challenges: measurement and identification.

The measurement issue arises because VMPL is unobserved, while welfare evaluation boils down to whether the policy reallocates factors to high-VMPL sectors and firms or not

9. We review the empirical and quantitative examinations of industrial policy, classifying them into two groups:

a) *ex post* evaluations using design-based methods such as DiD

b) *ex ante* model-based evaluations of optimally designed policies

8. At a deeper level trade openness presents a dilemma for industrial policy effectiveness—a point more recently emphasized in the literature.

7. Another important but often-overlooked point is that trade measures, such as import restrictions or export promotion, are rarely an optimal or efficient form of industrial policy.

6. Open economies face additional challenges when designing IP:

a. Trade introduces a *siphoning effect* where taxes paid by domestic households are partially used to subsidize consumption abroad via exports.

b. Industrial policy distorts relative trade prices ---> disrupts the gains from trade

5. An economy-wide *optimal* policy is rarely implemented in practice. More often, a a few sectors are promoted in isolation (e.g., South Korea’s HCI drive).

In such cases, the *constrained-optimal* policy depends on the upstreamness of the targeted industries.

4. The *welfare gains* from optimal policy, however, depend on which sectors are promoted.

If industrial policy promotes upstream sectors operating below their efficient scale, it would generate larger gains.

3. We formalize this argument by deriving the optimal policy in a unified framework with various distortions and input-output networks.

Contrary to common instinct, the optimal policy is *newtrok-blind* and *demand-blind*. It only depends on the size of the distortion wedge.

2. We first outline the welfarist rationale for industrial policy: with externalities or market failures, the *value marginal product of labor (VMPL)* is different across sectors

---> Welfare can be improved by reallocating resources from low-VMPL to high-VMPL sectors.