THE FED IS OFF ITS OWN REACTION FUNCTION. IT SHOULD TELL US WHY.

The following information was released by the Peterson Institute for International Economics:

Jason Furman (PIIE)

Date

October 27, 2026 2:45 PM

Body

Welcome to the first occasional post from me, Jason, a human. Almost everything on Econvitals is an automated briefing, but these posts will all be written by me, as are the chart(s) of the day and the opening paragraph of the Morning Note. I'm not sure how often I'll write these posts (thus the nonspecific "occasional"). My plan is to make them always data oriented but divided between the descriptive and the explanatory with a touch of opinion. This will not be the place for my broader thoughts on issues ranging from the California wealth tax to the latest proposals from President Trumpyou can find me opining on those elsewhere. Also, full disclosure: these posts are all copyedited and fact-checked by AI, and all the data analysis is done interactively with AI. Oh, and I suck at titles, so the AI may do those as well.

And now to substance. Four times a year the Federal Reserve's Federal Open Market Committee (FOMC) releases a Summary of Economic Projections (SEP) with its participants' expectations for growth, unemployment, inflation andmost notablythe federal funds rate (the so-called dots). The projections and the Fed's actual policy moves have gotten criticism in equal measure for the ways in which policy and the economy deviate from what the Fed sets out, raising questions about the value of forecasts that are often wrong and actions that often differ from what people interpret as the Fed's plans.

The Fed's forecasts[1] have, however, actually been a very good guide to its reaction function. It often ends up raising rates by less than the dots said it wouldbut in most of those cases that was the "correct" response to inflation coming in lower than it expected or unemployment higher. And the reverse when it raises rates by more than the dots said. At least this is what it did until the last few years, when the Fed has deviated from its implied reaction function by more than at almost any time since it started publishing interest rate projections in 2012 (pandemic excluded). That raises a question: is the Fed badly off course, or is there something different, and more obviously transitory, about this inflation? I actually think the Fed has a good answer to why it is so far off its reaction functionbut it should be more explicit about giving us that answer.

Now, more numerical detail on all of this.

The Fed's Economic Projections as a Reaction Function

The Fed is not a particularly accurate predictor of its own interest rate moves. Since 2012 it has published 51 sets of projections that can be scored against the set published a year later. Of those, the Fed ended up with higher-than-projected rates 15 times and lower-than-projected rates 21 times (the other 15 were mostly the zero-rate years, when the plan was no change and that is what happened). Some of the misses have been quite large: in December 2021 it projected three quarter-point hikes for 2022 and delivered 4.25 percentage points, and in December 2018 it projected two hikes for 2019 and delivered three cuts.

The question is not whether the Fed knows what it will do in the future, but whether it knows how it will react when the data depart from the expectations on which those future moves were conditioned. The Fed's forecasts of what will happen to the economy are at least as good if not a bit better than others' but very far from omniscient. So you can think of the projections not as a plan for its actions but as setting out a reaction function along the lines of "here is how much we expect to change interest rates, but if unemployment or inflation turn out different from our forecast we will adjust accordingly."

Take two examples:

In December 2018 the median FOMC projection showed a 0.5 percentage point increase in the federal funds rate over the next yeartwo quarter-point hikes. The Fed ended up cutting three times, by 0.75 percentage point, a swing of about 1.25 percentage points relative to its plan. That swing was fully consistent with its implicit reaction function. By December 2019 the Fed's own estimate of 2019 core PCE inflation was 0.4 percentage point lower than it had forecast a year earlier (1.6 percent instead of 2.0), and the unemployment rate 0.1 point higher (3.6 instead of 3.5)on top of which it had marked down its estimate of the natural rate of unemployment by 0.3 point and its estimate of the neutral federal funds rate by 0.3 point. Run through the original Taylor rule1.5 times the inflation miss, plus 1 times the miss in the unemployment gap, plus the change in the neutral ratethat is 0.6 + 0.4 + 0.3 = 1.3 percentage points, the adjustment the Fed made, to within rounding.

In December 2021 the Fed projected it would raise rates by 0.75 percentage point over the next yearthree quarter-point hikes. It ended up raising them by 4.25 percentage points. Most of the 3.5 percentage point overage was the Taylor rule adjustment: by December 2022 the Fed's estimate of 2022 core PCE inflation was 2.1 percentage points above its forecast (4.8 percent instead of 2.7), which calls for 1.5 × 2.1 = 3.15 percentage points of extra tightening, slightly offset by unemployment coming in 0.2 point higher than forecast (3.7 instead of 3.5). Net, the rule accounts for about 3 percentage points of the 3.5 point overage.

These two examples are typical. Figure 1 shows, for every set of projections since 2012 paired with the set twelve months later, how much a Taylor rule reaction function said the Fed should adjust its planned year-end rate given how its forecasts changed (horizontal axis), and how much it actually adjusted it (vertical axis). (The rule-implied adjustment also includes the Fed's revisions to its estimates of the longer-run unemployment rate and the neutral interest rate.) The line is the Taylor rule itself: on it, the Fed did exactly what the rule implies. The points cluster around it. If the weights are estimated freely, the best fit puts a coefficient of 1.57 on the core inflation revision and 0.99 on the unemployment revision, as compared with 1.5 and 1 in the Taylor rule. The fit excludes the pandemic rounds and the years when the federal funds rate was stuck at zero, when the Fed might well have wanted a lower rate if that were possible.

It turns out this reaction functiona Taylor rule in changes, applied to the Fed's own forecast revisionsfits the Fed's behavior better than any empirical Taylor rule estimated over the same period (some details and comparisons are in a Claude-written, Jason-edited online appendix to this post). And it fits reasonably well: the rule explains about three-quarters of the variation in how the Fed moves its year-end rate projections (an R of 0.77), with a typical miss of 0.66 percentage point.[2]

Which brings us to the present.

The Federal Funds Rate Is Currently Well Below What the Fed's Implicit Reaction Function Calls For

In December 2025 the FOMC expected to cut rates by about 25 basis points over the following year. The economy did not turn out as expected. The Fed had expected the unemployment rate to end 2026 at 4.4 percent; its most recent projection, from September, is 4.1 percent, enough under a Taylor reaction function to eliminate the planned cut.

The bigger, and less pleasant, surprise has been inflation: in December the Fed expected core PCE inflation of 2.5 percent in 2026, and its September projection is 3.4 percent. Under a Taylor rule reaction function that revision alone would justify a 1.35 percentage point hike in the federal funds rate (and the Fed's upward revision to its estimate of the longer-run neutral rate, from 3.0 to 3.2 percent, adds another 0.2). That is way more than the 0.25 percentage point hike delivered so far, or even the 0.5 percentage point cumulative increase for 2026 that the market is currently pricing in. (For what it's worth, my own unemployment forecast is slightly higher than the Fed's September one and my core PCE forecast slightly lower, so under my own forecast I would get a somewhat smaller increase.)

Overall, using the full method behind Figure 1, the Fed's December 2025 projections treated as a reaction function say it should have raised its planned end-2026 rate by 1.85 percentage points, to about 5.25 percent. It raised it by 0.7 relative to its original plans, to 4.1 percentroughly where the market is too. That shortfall of 1.15 percentage points is unusually large. Of the 37 comparable twelve-month readings since the Fed began publishing its rate projections in 2012 (leaving aside the pandemic and the zero-rate years), only five have been more dovish, and all five came in the past three years.

Moreover, this is not an artifact of one forecast. The Fed has published projections for the end-2026 federal funds rate in 13 rounds, starting in September 2023. Treat each of the twelve earlier ones as a reaction function, apply the Taylor coefficients to how its forecasts have changed since, and nearly all of them imply a substantially higher rateon average about 1.3 percentage points above the Fed's own current projection of 4.1 percent for the end of this year, as shown in Figure 2. (The exception is the June 2026 round, by which time most of the inflation revision had already happened.) A live version of Figure 2 is maintained on the Econvitals Taylor Rule page.

This is not just about the current Chair, Kevin Warsh. Use the March 2026 projections, the last produced under Chair Powell, as the anchor and the gap is smaller but still substantial: the rule says the Fed should have raised its end-2026 projection by 1.45 percentage points since March, and it raised it by 0.7, a shortfall of 0.75 percentage point.

The Fed Being Off Its Reaction Function May Be More a Sign of Wisdom Than Folly

One perspective is that the Fed's implicit reaction function is the right way to think about policy. The economy has been served well by Taylor-type rules, and years like 2021, when the Fed ended up far off its rule, ended up imperiling the U.S. economy. From this perspective, a deviation that ranks among the half dozen most dovish readings of the Fed's own record since 2012 is a flashing warning sign.

Another interpretation is that the Fed is wiser than the formula. It understands, for example, that some of the inflation is the transitory result of tariffs and the Iran conflictand that, at worst, these will raise the price level once and then stop adding to inflation. The Fed probably should not react to the inflation that results from an increase in a value-added tax; from that perspective, it should not react to tariff-induced inflation either.[3]

Which is it, folly or wisdom? Quantification can shed some light. Using the Fed's December 2025 forecast as the anchor, we can ask what forecast revisions would be consistent with its switching from a 25 basis point cut to the 50 basis points of hikes the market now expects this year. The answer: an unemployment rate of 4.2 percent combined with core PCE inflation of about 2.7 percent. That is roughly half a point below a reasonable current forecast of year-end core PCE inflation (about 3.2 percent)a gap that could reasonably be attributed to transitory factors like tariffs and the Iran conflict. And if the part of the inflation surprise that shows up in core PCE but not in the trimmed-mean or median measures is treated as noise, the Fed is roughly on its reaction function (the appendix shows the arithmetic).

I confess, I began this exercise thinking the Fed was way off basebut after quantifying it I end up more sympathetic to its approach.

The Bottom Line: Reaction Functions Are a Useful Communications Deviceand the Fed Should Explain Its Deviations From Them

I would like to see the Fed publish the Taylor rule recommendation at every one of its rate-setting meetings instead of just twice a year in its semiannual Monetary Policy Report, ideally picking one as its main reference rule. I like a Taylor-style reaction function even better where the Fed articulates a path and explains that it will deviate from it based on how inflation and unemployment deviate from its forecast. The latter works better when the levels change over time, although it also risks perpetuating and compounding errors over time.

In either case, however, the people on the FOMCand the staff who support themknow a lot more than the formula knows, and they should use that information in setting policy. But they should also clearly convey it. In the current situation that could take the form: "We were planning one 25 basis point rate cut this year; instead we are now on track for two increases. This is because the unemployment rate has been lower than we expected and inflation higher. Normally we would have raised rates by more based on an inflation forecast error of this size, but we are discounting about half a percentage point of it as due to transitory factors that we expect to unwind or reverse over the coming year. If we are wrong in that expectation we will make up for it with further rate hikes."

Appendix by Claude (Directed and Edited by Jason)

The reaction function in this post differs from a textbook Taylor rule in two ways: it is written in changes rather than levels, and it is run on the Fed's own forecasts rather than on the realized data. This appendix compares it with the familiar alternatives over the same periodJanuary 2012, the first set of projections with dots, through September 2026leaving out the pandemic rounds and the meetings at which the federal funds rate was stuck at zero (201215 and 202021), since no rule could be followed downward then. The underlying work is written up in an Econvitals research note, Fed minus rule, rolling, which gives the full method, the real-time data vintages and the longer history behind the Econvitals Taylor Rule page.

Table A1 asks the same question of each version: when the Fed moved, did the rule say it would? The forecast-revision rule explains about three-quarters of the variation in how the committee moves its year-end rate projection over twelve months, with a typical miss of 0.66 percentage point, against 1.22 for assuming the projection never moves. Giving unemployment the balanced-approach weight of 2 instead of 1 fits worse on every sample.

The same rule in changes run on the realized datathe change over twelve months in core PCE inflation and in the unemployment rate, as published at the time or as revised sinceexplains nothing over this period: its miss is larger than assuming the funds rate never changes. That is not because the Fed ignored the data. Policy since 2012 has been about levels: lifting off from zero on a schedule in 201518, catching up by more than four points in 202223 while trailing core inflation was already falling, then normalizing from a restrictive level. In 202223, for example, the Fed raised rates by one to four and a half points over twelve-month windows in which trailing core inflation was already falling and a rule in changes was asking for cuts; in 201617 it hiked on a schedule while the rule asked for a quarter point or less. Leave out the catch-up pairs and the realized-data rule explains about 30 percent of the Fed's moves, which load on the change in unemployment rather than on trailing inflation.

Pandemic pairs (ending in 202021) and pairs at the zero floor excluded. R is from a regression of the Fed's change on the rule's change; the miss is the root-mean-square difference between the two with no fitted slope or intercept. The realized-data rule uses the unemployment rate and 12-month core PCE inflation released by each meeting day (ALFRED vintages for "as known at the time"), every scheduled meeting paired with the one nearest 12 months later. Source: Federal Reserve Board; BEA; BLS; Econvitals calculations.

Table A2 asks the level question instead: meeting by meeting, how far has the funds rate sat from where each rule says it should be? Every Taylor-type level rule has put the rate below the rule at the great majority of meetings since 2012, by 1.3 to 2.7 percentage points on average, depending mostly on which neutral real rate is usedTaylor's original 2 percent gives the largest gaps, the committee's own longer-run estimate and a real-time LaubachWilliams estimate the smallestand only a fifth to a half of meetings fall within a point of any of them. A freely fitted level rule over the period (the funds rate regressed on core inflation and the unemployment gap) explains only about a third of the variation, with weights of roughly 0.6 on each, far from Taylor's. The two rules that do track the rate, the Board's inertial and first-difference rules, do so by construction: each starts from the current rate and moves a little, so they describe the rate's persistence rather than its level. By any of these level measures today's rate is below the rule by about as much as is normal for the period (the Taylor (1993) gap of 1.5 points is the 38th percentile of meetings since 1990), which is why this post's claim is about the Fed's own reaction function rather than about the Taylor rule in levels.

Core PCE inflation throughout; the unemployment gap uses CBO's natural rate unless stated. "Today's vintage" data except the real-time row. The rate compared is the target midpoint set at the meeting. Rule definitions are the Federal Reserve Board's. Source: Federal Reserve Board; BEA; BLS; CBO; Federal Reserve Bank of New York; Econvitals calculations.

Finally, the weights (Table A3). Fitted freely, the committee's response to its own forecast revisions has Taylor's shape: 1.57 on the inflation revision and 0.99 on the unemployment revision at twelve months, against the rule's 1.5 and 1. Fitted on the realized data over the same years, the weights are 2.4 on the change in unemployment and a negative number on the change in trailing core inflation; since 1990 they are 2.1 and about zero. The committee reacts to its inflation forecast, not to the trailing twelve months, and the SEP is where that reaction is visible.

Weights on the change in (or, in the last row, the level of) core PCE inflation and of the unemployment gap, signed so that a positive unemployment weight means a higher rate when unemployment is lower. The forecast-revision fits also include the revision to the longer-run funds rate, whose fitted weight is near zero. Source: Econvitals calculations.

Table A4 varies the inflation measure. Headline PCE revisions are a little more correlated with the dots than core revisions are, but the committee moves only about two-thirds as far as they imply, so the rule's own miss is larger and the fitted inflation weight falls to about 1: the committee looks through food and energy. The last two rows take that idea one step further, as a rough adjustment for supply shocks concentrated in a few categories. They subtract from the core PCE forecast revision the actual change, between the two projection dates, in the gap between 12-month core PCE inflation and the Dallas Fed's trimmed-mean or the Cleveland Fed's median PCE measure. Both adjustments raise the R slightly and the miss somewhat; neither fits as tightly as the base model. But they change the current reading a great deal. Between July 2025 and July 2026 core PCE inflation rose from 2.8 to 3.0 percent while the trimmed mean fell from 2.8 to 2.3 and the median from 3.5 to 2.7, so the adjusted inflation revision for the latest pair is close to zero and the Fed sits almost exactly on its reaction function. On the December 2025 anchor used in the text, the rule's 1.85 becomes 0.9 with the trimmed-mean adjustment and 0.6 with the median, against the 0.7 the Fed did.

Same 37 pairs as Table A1, Taylor weights (1.5 on inflation, 1 on the unemployment gap, the longer-run rate passed through). The adjustments use today's vintage of 12-month core PCE, the Dallas Fed trimmed-mean PCE and the Cleveland Fed median PCE, read for the month each set of projections had in hand. Source: Federal Reserve Board; BEA; Federal Reserve Banks of Dallas and Cleveland; Econvitals calculations.

Caveats: the medians for inflation, unemployment and the funds rate come from different participants, and each participant's forecast is conditional on his or her own policy path; the twelve-month pairs overlap, so the fit statistics are in-sample descriptions rather than forecasts; and 201226 away from the floor is 37 projection pairs and 70 meetings, a short record.

Notes

1. This post uses the shorthand "the Fed's forecasts" when all we actually see is the median of the FOMC participants' projections. The median for one variable can come from a different participant than the median for another, and the participants change over time. Still, I'll keep the language simple.

2. My preferred specification is core PCE inflation as a measure of the underlying inflation the Fed appears to react to, looking through the supply shocks to food and energy that make overall PCE inflation diverge from core PCE inflation. And, core PCE is more accurate in this exercise with a typical miss of 0.66 percentage point as compared to a typical miss of 0.81 when overall PCE inflation is used in the reaction function.

3. One rationalization of the Fed's recent actions is that it is looking through supply shocks, as it has generally done in the past, but is adopting a wider concept of them than just food and energy. The trimmed mean PCE inflation rate, a measure that excludes large positive and negative moves, has fallen by about 0.7 percentage point over the last year, to 2.2 percent, and the Cleveland Fed's median PCE inflation rate by about 0.9 point, to 2.6 percent, even as 12-month core PCE inflation rose about 0.2 point, to 3.0 percent, and the Fed's projection for 2026 core PCE inflation rose 0.9 point. Treat the widening gap between core and these measures as noise and the Fed is roughly on its reaction function; the appendix has the arithmetic.