The Fed's Five Task Forces: What We Recommend

We have written seven posts that address Chair Warsh's five task forces. The introductory post with links to all the others is here. This final post gathers the recommendations in a single place — 22 of them — and notes briefly how they fit together.

Our diagnoses share a common observation. In each of the five areas, the Committee steers by a number nobody observes: the inflation trend, the shock decomposition, the “stars”, the inflection point of reserve demand, and the reaction function itself. The Committee (or the public) infers these, typically using filters calibrated to a past distribution of shocks. Based on some combination of statistical filters and stylized models, these guides become fragile in key episodes, such as the broad pickup of prices in 2021.

The recommendations respond to this unobservables problem in five ways. In some, we ask the Fed to disclose what goes into the numbers it already produces. In others, we urge it to collect what it cannot currently see. In still others, we ask it to design policy that holds up when estimates prove wrong. In another, we urge it to keep checking that its routine estimation methods still work. And in two, we ask it not to discard what it currently produces before a credible replacement is available.

Publish the inputs, not just the output

This is the recommendation that appears in every post, reached independently each time. The Fed publishes conclusions and withholds the reasoning that produced them. Seven of our twenty-two recommendations ask it to disclose more.

Publish the Summary of Economic Projections (SEP) matrix. Each participant's projected rate path should appear alongside that same participant's forecasts for inflation, unemployment, and growth. The dot plot currently shows where each participant would set rates. It does not show why.

Identify the SEP dots. Anonymity makes it impossible to see whose reading of the economy has held up and whose has not. The Committee will resist this recommendation most fiercely, which is why the Communications Task Force should address it directly.

Publish the reserve demand indicators, and the error record. The Committee judges reserve adequacy against a level it cannot observe: namely, the inflection point where the reserve demand curve steepens. It should highlight what it actually watches — the SOFR-IRB spread, the timing of dealer-bank payments, the take-up at standing liquidity facilities, and so on — and then publish that information.

Explain which inflation number does which job. The FOMC reads trimmed means, medians, and short-term averages internally. It reports headline inflation externally. It has never explained the relationship between them. The Bank of Canada has. Transparency will boost credibility.

Routinely publish the internal analytic methods. Disaggregation, trimming, and asymmetric treatment of the tails of the distribution of prices are routine judgment calls, and each can affect policy decisions. Yet, none appears in any FOMC statement. Publish them, along with the skew and the breadth of price increases, so that anyone can see when the filters used to estimate trend inflation stop working properly.

Publish the methods applied to private data, and negotiate the right to release summary statistics. If the FOMC comes to rely on numbers few outsiders can inspect, it will have traded one opacity for another.

Publish a reliability record for the “stars.” For potential output (y*) and the natural rates of interest (r*) and unemployment (u*), report the distribution of past revisions alongside each current estimate: the average error, the spread, and the episodes of serially correlated errors.

Collect what the Fed cannot now see

Transparency about existing numbers only goes so far. Some of what the Committee most needs, no government agency collects.

Measure prices and quantities together, and record what the goods are. Prices alone cannot separate a supply shock from a demand shock. Quantities supply the identification, and the Fed does not observe them directly: the Census Bureau gathers revenue, the BLS gathers prices, and the BEA derives the rest. Item-level transaction data deliver both at once. This is the single most valuable thing the Data Task Force could propose.

Buy the scanner data and build redundancy. A statutory mandate does not pause when the statistical agencies shut down for lack of funding. Private sources operated straight through the last shutdown. The Fed should have the scanner data in hand before the next one.

Track the realized economic impact of AI, not just expert beliefs about its progress. What nobody publishes is a running account of what has actually happened: unit costs where adoption is highest, hiring by occupational exposure, shifts in within-job activities and the like.

Regularize the diffusion measure the Fed already produces. The St. Louis Fed's AI adoption indicator should move to a regular schedule and be linked to outcomes. The Census Bureau separately measures adoption across 1.2 million businesses every two weeks. Nobody can yet link a firm's adoption to its prices, headcount, and output. That linkage is a project for Census, the BLS, and the BEA together — and the Fed is the most powerful customer these statistics have.

Commission the cost-benefit work on Fedwire netting and intraday overdrafts. These address the root cause of high structural reserve demand rather than its symptoms. Both projects will take years, and neither will be ready when the pressure to shrink the balance sheet arrives unless the work starts now.

Build a policy framework that survives being wrong

Better data will not eliminate uncertainty. Some estimates will remain unreliable no matter how much the Fed spends on measurement, so policy design has to account for that directly.

Commission work on alternatives to the “stars.” If the estimates of key benchmarks are unreliable for a decade at a stretch, the sensible response is to lean on them less rather than estimate them with more conviction. Difference rules are one candidate. Replacing a filtered benchmark with one built from observables is another.

Make the reaction function explicit. Without one, no framework can be credible. The FOMC should expand the Summary of Economic Projections (SEP), providing details needed to communicate its reaction function clearly.

Add scenario analysis. The matrix we propose traces the reaction function locally, showing how policy responds to small deviations around the central forecast. It says nothing about the tails. Scenarios fill that gap — and when the range of outcomes is as wide as it now is, the Committee should say in advance that its own path is uncertain rather than surprise markets later.

To ease reserve demand, start with temporary open market operations. Sterilizing the predictable supply shocks — quarter-end window dressing, Treasury General Account swings, movements in the FIMA pool — requires no legislation and no new authority. The Fed did this routinely before the crisis. It is the one tested reform available now to prepare the way for a smaller balance sheet.

Attenuate policy responses until AI uncertainty recedes, and be prepared to act aggressively when it does. A policymaker unsure of either an instrument's potency or of current conditions should move in smaller steps than certainty would justify. Caution preserves the option for aggressive action later.

Keep checking that the methods still work

Every rule for separating signal from noise reflects a particular distribution of shocks. When the distribution shifts, the rule silently stops working. Three recommendations address that.

Keep checking the filtering methods. The Dallas Fed built its (asymmetric) trimmed mean using data from an era of falling goods prices. Tariffs and other supply shocks have reversed that pattern. The distribution will keep shifting, and the filters must keep up.

Choose the external inflation measure and its target number together. Pick an index the public recognizes, then set the number consistent with that index and its average measurement bias. Track the bias as a check on the target. Chair Warsh has already committed to the 2 percent inflation target for now. Proposals for a higher target rested on limited policy space near the effective lower bound. Five years of above-target inflation have changed that calculus: the risks of upward inflation drift and lost anchoring now exceed the risk of insufficient headroom.

Apply makeup strategies only in exceptional circumstances, ensuring they are symmetric and explicit. Flexible average inflation targeting (FAIT) failed for a clear reason: the FOMC never specified key parameters such as the averaging window or the restoration period that would have constrained discretion.

Do not dismantle before the replacement works

Two recommendations run against the direction policy is currently heading. Both make the same point: do not remove something that works today on the promise that something better will arrive tomorrow.

Enrich the dot plot. Do not end it. A Committee that stops publishing its projections does not become more accurate. It becomes harder to hold accountable. The FOMC’s 2021 forecasting failure is visible today only because the SEP projections were published.

Sequence reserve demand before reserve supply. Measures that shift reserve demand inward must be in place, and shown to work, before the FOMC seeks to reduce reserve supply. The reverse order is how September 2019 happened. The Balance Sheet Policy Task Force should say plainly that political pressure to act prematurely poses unnecessary risks.

The recommendations by task force

Communications (post). Enrich the dot plot, do not end it. Publish the matrix. Identify the dots. Add scenario analysis.

Balance sheet policy (February, April). Reduce reserve demand before trimming supply. Start with temporary open market operations. Publish the indicators of reserve demand and the error record. Commission the research on Fedwire netting and intraday overdrafts that eventually could lower reserve demand substantially.

Inflation frameworks (here). Make the reaction function explicit. Avoid complex “makeup” provisions or frameworks that depend heavily on the long-term reliability of a price index. If a makeup provision is included, make it symmetric and sufficiently explicit to constrain discretion.

Data — measuring inflation (here). Distinguish between a simple public target and internal tools for estimating inflation’s trend. Explain which number does which job. Choose the external measure and its quantitative target number together. Routinely publish the internal analytic methods. Keep checking that the filtering methods still work.

Data — what to measure (here). Buy the scanner data that measure prices and quantities together. Build redundancy. Record the attributes of goods and services to improve quality adjustments. Measure the frequency at which firms change prices. Publish the analytic methods and negotiate the right to publish summary statistics of private data.

Productivity and jobs (here). Attenuate policy responses until AI uncertainty recedes, and be prepared to act aggressively when it does. Understand that the faster AI advances, the less reliable key Fed policy guides will become. Track the realized evidence of AI’s economic impact, not just the beliefs about AI’s progress. Regularize the FRB St Louis AI diffusion measure. Commission work on alternatives to the “stars” and publish a reliability record for the star estimates.

What this adds up to

None of these recommendations asks the Fed to forecast better. We do not think it can. The more interesting question is what to do amid heightened uncertainty.

What the task forces can do is shorten the interval between the world changing and the Committee noticing. Until then, they can push the Committee to consider policy approaches that hold up when its usual guides prove wrong. In a few cases, the task forces also can prod the Committee to gather information that might reveal errors early.

That is the opportunity in front of these five task forces. It is a large one, and it will require ambition.