Why Macroeconomic Policy Needs to Move from “Uniform Treatment” to “Verifiable Precision Treatment”
On September 16, the Federal Reserve raised rates by another 25 basis points, lifting the target range for the federal funds rate to 3.75%–4.00%. The rationale in the statement was straightforward: the U.S. economy was still expanding at a solid pace, domestic spending remained resilient, productivity was strong, capital investment was robust, employment was broadly stable, and inflation remained elevated. [1]
In his earlier Jackson Hole remarks, Kevin Warsh cited another persuasive figure: among 199 detailed components of the Personal Consumption Expenditures (PCE) price index, 54% had risen by more than 3% over the previous 12 months; over the previous six months, 49% had increased at an annualized rate above 3%. [2]
That number easily creates a particular impression:
Inflation is no longer a problem confined to oil, food, or a few isolated goods. Prices are rising across the economy.
If so, another rate increase seems almost self-evident. But the 54% figure is exactly where the article needs to be opened up. It proves that many prices are rising. It does not prove why they are rising, and still less that a uniform rate increase is the best treatment.
In the past, opening all 199 components one by one—examining their weights, pricing mechanisms, common shocks, and sensitivity to interest rates—was costly. Today an ordinary person, with the help of artificial intelligence (AI), can keep asking: What exactly is inside that 54%? Which components carry the most weight? Which reflect genuine demand overheating? Which come from oil, housing, health care, financial markets, or statistical methodology? And how many of those price increases can actually be restrained by higher interest rates?
Once that layer is opened, the problem changes.
1. If 54% of Components Are Rising, That Does Not Mean 54% of the Economy Is Overheating
Start with financial services. Fees for portfolio management and investment advice are often related to the value of assets under management. A study by the U.S. Bureau of Economic Analysis (BEA) notes that the current U.S. method for pricing some fund-management services incorporates part of the upward revaluation of asset prices into measured price increases. [3][4] In other words, a rising stock market can itself push up measured financial-services prices in PCE. That increase is real. But it is not the same as a restaurant raising prices because too many customers are competing for too few seats. The issue is not whether it belongs in PCE; the issue is whether it can be treated directly as evidence of overheated aggregate demand.
Health care is even more complicated. U.S. medical prices are not ordinary market prices.
A patient may pay only $20 for a doctor’s visit while the insurer pays another $80. When medical-service prices are measured, the relevant price is the $100 total reimbursement received by the provider, not merely the $20 paid out of pocket by the patient.
Health-insurance pricing is more unusual still. The U.S. Bureau of Labor Statistics (BLS) does not simply track how much the same insurance policy costs this year versus last year. It estimates the price indirectly from premiums, benefits paid, and insurers’ retained earnings, with some of the underlying information arriving with substantial lags. [5]
So rising medical PCE can come from higher hospital charges, renegotiated insurer-provider contracts, changes in Medicare payment rules, estimated insurance-service prices, or genuine demand for medical care. These are not the same cause.
Now consider housing. In August, U.S. existing-home sales fell 2.0% from the previous month and 1.2% from a year earlier. Inventory rose to 1.62 million units, equivalent to 4.9 months of supply. New-home inventory was equal to 8.5 months of sales, while the median new-home price was down 5.8% from a year earlier. [6][7]
It is difficult to describe that market as overheated. Yet housing-services prices in PCE can still show relatively strong increases. The reason is simple: housing PCE measures the price of housing services, not the real-time temperature of the property transaction market, and it moves with a lag. Housing PCE can therefore still look hot after the housing market itself has already turned cold.
Energy is similar. When oil prices rise, the effect does not remain inside the gasoline category. It passes into air travel, transport, delivery, chemicals, food, and many service costs. One oil-price shock can therefore appear as price increases across a dozen—or dozens—of PCE categories. Statistically, that looks like “diffusion.” Economically, many of those increases may still be multiple consequences of the same cause.
Warsh’s 54% figure is therefore real. But what it actually tells us is that many price categories are rising. It does not tell us whether those increases are all being generated by the same overheating of aggregate demand. Those are very different statements. There is another layer as well: the 54% is a count of categories, not a share of PCE expenditure weight. Even if 107 categories are rising, that does not mean 54% of consumer spending is overheating.
2. PCE Is Real, but PCE Is Not the Diagnosis
Medicine offers a simple analogy. In the past, when doctors did not know the cause of an illness, they saw fever, pain, and weakness and treated the symptoms—sometimes even with bloodletting. Then came microscopes, blood tests, X-rays, computed tomography (CT), magnetic resonance imaging, and genetic testing. Today, if two patients both have a temperature of 39°C, a doctor does not give them the same medicine merely because the thermometer shows the same number.
In July, headline PCE inflation was 3.7% year over year and core PCE inflation was 3.3%. Those figures are real. [8] But another picture exists at the same time. The Dallas Fed’s Trimmed Mean PCE inflation rate was only 2.3%, a measure specifically designed to reduce the influence of categories with unusually large price movements. [9] BEA’s market-based core PCE inflation rate was about 3.0%, also below ordinary core PCE; it attempts to exclude some consumption items that lack directly observed market prices and therefore require estimation. [10] These measures do not tell us which one is the “true” inflation rate. They tell us that the same economy can produce different inflation pictures when viewed with different rulers.
The real policy question therefore cannot stop at whether PCE is above 2%. It must ask why it is above 2%.
Financial services may be affected by asset prices. Medical prices include a large institutional and contractual component. Housing prices are lagged. Energy can reflect an external shock. Tariffs can spread through supply chains. AI capital spending, meanwhile, may genuinely be creating localized investment overheating. All of these end up inside PCE.
But they are not the same disease.
3. The U.S. Economy Is Not at One Temperature
If we open the U.S. economy by sector, we do not see an economy that is uniformly overheated.
A better description is: AI is hot; housing is cold; consumption is warm; employment is normal; and energy is being affected by external shocks.
When the economy contains so many different temperatures, adjusting everything through one policy rate creates a practical problem: the sectors hit first may not be the hottest sectors.
Large technology companies have cash flow, global bond markets, private credit, and the ability to issue dollar debt in Europe or Asia.
Ordinary households do not have those choices. Small and medium-sized businesses do not either. A 25-basis-point increase may be a relatively small variable for a highly profitable AI investment with many financing channels. But for a property project, a household mortgage, or a small business dependent on bank credit, it can determine whether the project is viable at all. The paradox is obvious: the Fed wants to cool the economy, but the first places to cool may not be the hottest ones.
4. Why Not Treat Sectors Separately?
The idea is not new. China has long used structural monetary-policy tools. Europe uses sectoral capital buffers and macroprudential tools such as loan-to-value (LTV) limits. Japan has used targeted funding-support tools. Russia adjusts restrictions dynamically for risks in housing and consumer credit. India has long used priority-sector lending. The United States itself has designed separate facilities for businesses, municipalities, small firms, and different credit markets during crises. The Basel Committee has argued that when risks are concentrated in one credit segment, capital requirements can be increased for that segment rather than tightening the entire financial system, thereby reducing the side effects of broad measures. [11][12] These instruments are not identical, but they share one feature: they do not rely on a single policy rate to solve every local problem.
America’s exceptionally capable economists have obviously thought about sector-specific treatment. At that point the issue is no longer purely economic. It becomes a question of fairness and power. Who has the authority to decide which sector should face a higher financing cost and which should face a lower one?
5. What the United States Really Worries About: Once Policy Becomes Precise, Who Guarantees Fairness?
A uniform interest rate is crude, but it has a major institutional advantage: formally, everyone faces the same benchmark price. Once policy starts differentiating by sector, the questions appear immediately.
Why should AI face an extra 150 basis points rather than 80? Why should housing receive a 100-basis-point reduction? Why should agriculture receive support but autos not? What exactly counts as an AI company? Who sets the standard? Who receives the benefit? Who bears the cost? When does the policy end? Does it become industrial policy? Does it create rent-seeking? One reason the United States has long preferred uniform market pricing is precisely that it is difficult to prove that a precise allocation is fair. Society may therefore prefer something crude but apparently even-handed, while remaining suspicious of something precise whose judgments are controlled by a small group.
6. What AI Really Changes Is the Cost of Verification
That is the genuinely new variable. In the past, analyzing all 199 PCE components one by one and then connecting them to bank loans, corporate bonds, private credit, housing inventories, electricity demand, profits, and returns on capital required enormous effort. Ordinary people could not do it. A small research team could not maintain it continuously in real time. Even if the Fed did the work internally, ordinary people could not verify it, and other experts would struggle to reproduce it. Complex reality therefore had to be compressed into a small number of aggregate indicators and then handed to an expert committee for judgment.
AI sharply lowers that cost for the first time. An ordinary person can now keep opening Warsh’s 54% figure layer by layer. With AI—and even more so with an intelligent agent—this is not an exhausting task. Each component can be checked, and AI can organize the conclusion, the evidence, and the counterarguments. The era has changed; the method now has the conditions to change with it.
In the past, the public usually had only two choices when facing experts:
Believe the experts. Or do not believe the experts.
Now there is a third choice: verify the experts.
That is one of the deepest ways AI can change public policy: it can make public policy open to verification.
7. Fairness Does Not Necessarily Mean Giving Everyone the Same Rule
Traditional systems easily equate fairness with exposing every sector to the same policy rate. But market conditions differ, and the same rate has very different real effects across sectors. Formal uniformity is not the same as fairness in outcome.
The AI era may make another concept of fairness possible: different conditions can justify different policies, but every difference must be supported by a public and verifiable reason.
Why was the policy triggered? What data support it? Why this magnitude? What evidence points the other way? What facts would show that the original judgment was wrong? When does the intervention end? These questions should all be open to public scrutiny.
Fairness does not require every case to be treated identically. It requires every difference in treatment to be explainable. AI cannot replace congressional authorization, central-bank independence, or legal responsibility. But it can change something that was extremely difficult in the past: after authority has been granted, differentiated judgments no longer need to remain inside a black box.
8. What Is Missing Is Not Structural Tools, but a Verification Layer
Many structural tools already exist around the world. The AI era therefore does not require inventing an entirely new macroeconomic-policy system. What is missing is an additional layer on top of the existing system: continuous verification.
The future architecture can be simple.
First layer: a uniform benchmark interest rate, used for genuine economy-wide aggregate-demand problems. Saying that aggregate tools have fallen behind does not mean they are useless. It means aggregate tools alone are no longer enough.
Second layer: structural financial tools, used for localized credit overheating, financial risk, and financing shortages.
Third layer: AI-supported verification, continuously asking: Is this sector really overheating? Why? Is the policy working? Is financing being rerouted? Is the intervention damaging other sectors? Has the original judgment failed? When should the policy exit?
Then targeted policy is no longer simply whatever experts decide.
It can have data, reasons, counterevidence, failure conditions, and an exit mechanism.
9. Warsh Needs to Open the 54% Figure One Layer Further
Warsh has already incorporated AI into his discussion of productivity, employment, capital formation, and potential output, and he has explicitly called for the Federal Reserve to obtain more accurate, timely, and actionable data. His observation that 54% of 199 PCE components had risen by more than 3% is already an attempt to move beyond looking only at aggregate PCE. [2]
But one more step is needed. The 54% merely breaks one aggregate indicator into 199 smaller indicators. To enter the AI era in a meaningful way, the analysis must continue downward:
Why did each price rise? What is its weight? Does it reflect a common shock? Is it a directly observed market price? Is it an administered or institutional price? Is there a statistical lag? Is it sensitive to interest rates? Which tool is appropriate?
Without that next step, we have only replaced one thermometer with 199 thermometers. We still have not found the disease. This is exactly where AI can help most.
10. From Disclosure Transparency to Verifiable Transparency
Traditionally, central-bank transparency has meant publishing statements, explaining decisions, releasing minutes, and publishing forecasts. All of that matters. But the AI era can go one step further:
Allow anyone to open a policy judgment layer by layer and inspect the evidence.
What are the facts? How was the indicator constructed? Which parts are facts and which parts are judgment? What alternative explanations exist? Why was this tool chosen? What facts would overturn the judgment? What happened after implementation? This is not simply more information. It is a different kind of transparency: verifiable transparency.
In the past, complexity itself was part of the expert barrier. AI is now lowering that barrier.
The next meaningful change at the Federal Reserve should therefore not merely be using AI to forecast PCE, nor merely studying how much AI will raise productivity. It should be making the policy judgment itself verifiable.
Conclusion
The global role of the dollar means that Federal Reserve policy spills over into almost every major economy. One reason other central banks have developed structural tools is that they need ways to manage local problems beyond the uniform policy rate. The United States has the same need. When the economy has become highly differentiated and diagnosis can become increasingly granular, must the Fed still rely primarily on one interest-rate hammer?
What the AI era should bring is not merely a smarter central bank, but a new form of macroeconomic management:
Aggregate problems should be handled with aggregate tools.
Structural problems should be handled with structural tools.
Every differentiated allocation should be open to public verification.
In the past we may have had to choose between two systems: crude, but apparently fair; or precise, but difficult to prove fair.
AI offers a third possibility for the first time: precision that is also verifiably fair.
Structural tools may offer the United States another benefit as well: separating the aggregate interest rate from sector-specific financial conditions. Suppose aggregate inflation eventually falls far enough to permit lower benchmark rates while parts of the AI supply chain remain locally overheated. The Fed could lower the benchmark rate while using capital requirements, risk weights, or other structural tools to raise the effective financing cost of highly leveraged AI infrastructure. Such sector guidance need not be permanent; it could be updated monthly or quarterly with explicit trigger and exit conditions. That raises another result worth testing: if the United States can combine a lower aggregate cost of capital with more targeted risk pricing, global dollar capital may become more willing to return to U.S. assets, potentially reinforcing the dollar’s central role.
This article itself was written by an ordinary person using verifiable thinking in collaboration with AI. If an ordinary person can use AI to take Warsh’s 54% figure all the way down to causes, tools, and fairness—and then place the result before experts for public scrutiny—that fact alone shows that the era has changed.
References
[1] Federal Reserve Board, “Federal Reserve issues FOMC statement,” September 16, 2026.
[2] Kevin Warsh, “Keynote Remarks at the 2026 Jackson Hole Economic Policy Symposium,” Federal Reserve Board, August 28, 2026.
[3] U.S. Bureau of Labor Statistics / Federal Reserve Bank of St. Louis, “Producer Price Index by Industry: Portfolio Management and Investment Advice,” July 2026.
[4] Robert Kornfeld, “Alternative Price and Volume Measures for Commercial Bank and Fund Management Services,” U.S. Bureau of Economic Analysis, 2021.
[5] U.S. Bureau of Labor Statistics, “Measuring Price Change in the CPI: Medical Care,” methodology for medical services and health-insurance pricing.
[6] National Association of Realtors, “Existing-Home Sales Report Shows 2.0% Decrease in August,” September 10, 2026.
[7] U.S. Census Bureau and U.S. Department of Housing and Urban Development, “New Residential Sales, August 2026,” September 24, 2026.
[8] U.S. Bureau of Economic Analysis, “Personal Income and Outlays, July 2026,” August 26, 2026.
[9] Federal Reserve Bank of Dallas, “Trimmed Mean PCE Inflation Rate,” July 2026.
[10] U.S. Bureau of Economic Analysis / Federal Reserve Bank of St. Louis, “Market-based PCE excluding food and energy,” July 2026.
[11] Basel Committee on Banking Supervision, “Guiding Principles for the Operationalisation of a Sectoral Countercyclical Capital Buffer,” 2019.
[12] Basel Committee on Banking Supervision, “Countercyclical Capital Buffers,” Basel Framework; discussion of sector-specific credit cycles and targeted tools.
