There is a tempting symmetry between two of the most-cited problems in macroeconomic policy. On one side stands Kydland and Prescott’s 1977 paper on time inconsistency — a government that announces optimal policy, watches expectations form, and then finds it locally rational to deviate, so the announcement is never credible to begin with.1 On the other side stands the much older Mill-Bastable problem in trade policy — a temporary tariff is supposed to nurture an infant industry until it can compete unprotected, but the tariff is never temporary because the political moment to remove it never arrives. Both are commitment problems. Both have the same logical structure. Both have generated decades of attempts to design institutions that bind the policymaker to her announcement.

It is tempting, on the strength of that isomorphism, to look at the central bank — the most successful commitment device the modern state has built — and ask whether the same architecture can be ported to industrial policy. An independent CHIPS Committee, a statutory mandate to phase out subsidies on a fixed timetable, members appointed for long terms and insulated from the next election. The recent revival of industrial policy in the United States makes this an almost irresistible move. Réka Juhász and Nathan Lane, in their 2024 NBER paper The Political Economy of Industrial Policy, frame the problem in exactly the Kydland-Prescott form: when government commitment is absent, firms expecting indefinite protection underinvest in cost reduction, and infant-industry policy becomes counterproductive.2

I think the isomorphism is real at the abstract level and broken at the implementation level, and the gap is exactly where you have to look.

Where the isomorphism holds

Strip both problems to their bones and they map cleanly. There is a policymaker with optimal announcement, a private sector forming expectations, and a future date at which the policymaker faces an ex-post temptation to renege. Rational anticipation of that temptation collapses the announcement back into the discretionary equilibrium. Both problems are solved, in principle, by some apparatus that takes ex-post discretion away — what Kydland and Prescott called rules rather than discretion.1 For monetary policy, the canonical implementation is the independent central bank with a statutory inflation mandate. For industrial policy, the proposed implementations include sunset clauses, independent appropriations review, and discipline mechanisms tied to performance.

So far, so good. The trouble is that the central bank’s institutional design works only because it sits on top of three structural features that monetary policy happens to have and industrial policy happens to lack. Pull any of the three away and the architecture stops being a commitment device.

The first condition: a single observable

Modern inflation targeting works because there is a number that everyone agrees on. Headline CPI, core CPI, the PCE deflator — these differ at the margins, but they are public, measured by an arms-length statistical agency, and disputes about them are technical rather than political. A central bank can be evaluated against a target like two percent without anyone having to argue, in any given quarter, about what “doing well” means.

Industrial policy has no such number. Productivity gain, export share, employment quality, supply-chain resilience, national-security capacity, knowledge spillovers — all of these are plausible objectives, all of them are in tension with each other, and all of them require political weighting before they can be measured. An “independent” industrial policy committee that has to choose among them is not actually independent; it has just relocated the political choice inside the chamber where the technocrats sit. You cannot delegate around a target that does not exist.

Korea’s Economic Planning Board, which operated from 1961 to 1994 and is often invoked as the developmental-state success story, is partly an exception that proves the rule.3 Park Chung-hee’s regime invented a single observable — export performance, priced in international markets — and welded subsidy continuation to it. That construction is what made discipline possible. The construction was political, contingent, and not easily exported; “make up a single observable” is not a generally available move for a democratic government negotiating with three semiconductor firms about where to build their fabs.

The second condition: reversible policy variables

The central bank moves a flow variable. The overnight rate goes up twenty-five basis points this month and can come down twenty-five basis points next month, with the cost of the round trip largely absorbed by financial markets in days. Expectations can be reset; Volcker rebuilt the Fed’s anti-inflation credibility within a few painful years precisely because the policy instrument was reversible even if the political costs were not.

Industrial policy moves stock variables. A fabrication plant, once sited, is not a setting on a dashboard; it is concrete, supply contracts, ten thousand jobs, and a regional political constituency. The plant cannot be unsited next month if the policy proves wrong, and the sunk costs accumulate into a political bloc that will be perfectly willing to lobby against the sunset clause when it arrives. The Kydland-Prescott solution — bind yourself to a rule and reset expectations when you fail — assumes that the underlying instrument is something like a dial. When the instrument is a factory and a supply chain, the dial metaphor stops working.

This asymmetry between flow and stock is, I think, the asymmetry most often glossed over in discussions of industrial-policy commitment. It is not just that exit is politically harder; it is that the policy variable, considered as a physical and contractual object, is not the kind of thing that can be returned to its prior state.

The third condition: atomized counterparties

The central bank confronts millions of price-setters and asset-holders, none of whom can sit across a negotiating table from the chair. This atomization is not incidental; it is the structural protection against capture. There is no individual wage-bargainer whom the Fed can be lobbied to favor, because the Fed does not interact with individual wage-bargainers at all.

Industrial policy confronts a handful of named firms. In semiconductors, you can count the relevant counterparties on one hand: TSMC, Intel, Samsung, perhaps Micron and SK Hynix. An “independent” CHIPS committee that has to allocate billions of dollars among these firms is, by construction, in bilateral negotiation with each of them. The classic anti-capture move — make the regulator face only diffuse interests — is unavailable. Dani Rodrik’s framing of “embedded autonomy,” in which government stays close to industry without being captured, is the honest description of the best case, but it is also an admission that the central-bank architecture is not what is on offer here.

What the CHIPS Act actually faces

The U.S. CHIPS and Science Act appropriated about $52.7 billion for semiconductor manufacturing, R&D, and workforce development, with roughly $39 billion of that for manufacturing incentives and additional investment tax credits layered on top.4 A working group at the next administration can, in principle, claw subsidies back, renegotiate the terms with TSMC’s Arizona facility, or hold up future tranches. The fact that some such intervention has reportedly been discussed under the current administration is exactly the time-inconsistency problem manifesting in real time.

The natural response is to ask why Congress does not bind itself with an “independent CHIPS authority,” some structural cousin of the Federal Reserve. The answer, when you look hard at the three conditions, is that none of them hold. There is no inflation rate of industrial policy. The factory at Phoenix is not a flow variable. The counterparty is not atomized. Whatever can be borrowed from central-bank design — narrow mandates, long terms, fixed review cycles — does not add up to the central-bank architecture, because the underlying structure is wrong.

What can be ported, and what to do instead

There are pieces that do travel. Narrow-mandate R&D agencies of the DARPA, ARIA, and Sprind family inherit the central bank’s “small mission, high insulation” design and seem to function, though with the important caveat that their remit is to fund research rather than to allocate production. Sunset clauses, written cleanly into appropriations, can perhaps stiffen credibility at the margin, though the empirical case for their effectiveness in industrial policy is still thin.

Beyond that, the realistic options look less like central banks and more like the messy, condition-specific arrangements that have actually delivered industrialization in the past — Korea’s manufactured single observable, Japan’s informal MITI coordination with industry, Singapore’s strategy of running the state as a single firm. None of these look like the Federal Reserve. Each of them depends on conditions that are not generally reproducible.

What this leaves me with, when I look across the whole pattern, is a methodological lesson rather than a policy recipe. Whenever someone proposes to solve a commitment problem by copying an architecture that worked elsewhere, the question to ask is not whether the two problems are abstractly isomorphic. The question is whether the three conditions that made the architecture work — a single agreed observable, a reversible policy variable, atomized counterparties — also hold in the new domain. If any of them is missing, the architecture will not survive the transfer. It will look like the original from the outside, and behave like nothing in particular from the inside.

The question worth asking about CHIPS, then, is not “can we build an independent committee” but “do we have anything that plays the role of inflation, the overnight rate, and the millions of wage-bargainers?” When the honest answer is no, the next move is not to keep stretching the central-bank template. The next move is to look harder at the architectures that fit the conditions that actually obtain.

  1. Kydland, Finn E. and Edward C. Prescott. “Rules Rather than Discretion: The Inconsistency of Optimal Plans.” Journal of Political Economy, vol. 85, no. 3, 1977, pp. 473-491. The 2004 Nobel Memorial Prize in Economic Sciences was awarded to Kydland and Prescott in part for this contribution; see the Royal Swedish Academy of Sciences’ advanced information document. Accessed 2026-05-19. ↩ ↩2

  2. Juhász, Réka and Nathan Lane. “The Political Economy of Industrial Policy.” NBER Working Paper No. 32507, May 2024; published in Journal of Economic Perspectives, vol. 38, no. 4, Fall 2024, pp. 27-54. Accessed 2026-05-19. ↩

  3. For background on the Korean Economic Planning Board (1961-1994) and its role in coordinating export-oriented growth, see the chapter “Korea’s Evolving Business-Government Relationship” in The Practice of Industrial Policy: Government-Business Coordination in Africa and East Asia (Oxford Academic). Accessed 2026-05-19. ↩

  4. Pearl Cohen. “CHIPS and Science Act Eligibility Guidance: $52.7 Billion in Funding and Tax Credit Incentives.” Accessed 2026-05-19. See also the Wikipedia overview of the CHIPS and Science Act for funding breakdown ($39B manufacturing, $13B R&D and workforce, plus ~$24B in investment tax credits). Accessed 2026-05-19. ↩