Welcome to the sixth issue of the Knightian Uncertainty Dispatch — a monthly curated reading list for anyone thinking about macroeconomics, finance, and forecasting in a world where the future is not just risky, but sometimes genuinely unforeseeable because it may differ from the past.

Each month, I recommend four new papers + one “essential” that help grapple with structural change and uncertainty beyond probabilistic risk — and what those realities imply for economic modeling, forecasting, and policymaking.

The goal is to curate papers that:

1. Deepen our understanding of market outcomes and policy in a world with unforeseeable structural change and Knightian uncertainty.

2. Are useful for how we actually reason and forecast in unstable environments.

3. Connect to one another — so the pieces speak to each other rather than living in isolated silos.

This month’s theme is shifting forecast errors.

The April Dispatch and the posts behind it documented the institutional shift now underway in central-bank forecast communication: scenarios replacing fan charts, no designated central projection, robustness across scenarios written into the analytical framework of major central banks. That shift is a response to a problem that has been visible in the data for forty years and that the post-pandemic experience has made impossible to overlook. Forecasts produced by major institutions and market participants deviate from outcomes in systematic, regime-contingent ways. The deviations are not noise. They have structure. And the structure is not what the standard interpretations predict.

This issue is built around two institutional documents and two academic papers that, taken together, document the structure of the deviations clearly. The Sveriges Riksbank published its statutory Account of Monetary Policy 2025 on March 3. Two months earlier, Morten Ravn and Carolyn Wilkins delivered the fifth in a series of external evaluations of Swedish monetary policy that the Riksdag commissions roughly every five years, covering 2015–2024. Together, the two documents provide a public record of a major central bank’s forecast performance through the post-pandemic inflation surge and its aftermath. Stillwagon (2026) applies Bai-Perron structural break tests to the Coibion–Gorodnichenko forecast-error regression on US Survey of Professional Forecasters data; once breaks are allowed for, the constant-parameter regression’s conclusions reverse. Granziera, Jalasjoki, and Paloviita (2025) document the threshold-conditional form of the same finding for the ECB’s inflation forecasts — forecast bias is concentrated in regimes where inflation exceeds 1.8 percent and is invisible to averaged tests.

This issue includes two essentials: Zarnowitz (1984) and Lovell (1986) — the foundational empirical case, made when much of the profession doubted survey data could test the theory at all, that survey forecasts reject the rational expectations hypothesis and its prediction that forecast errors should be unpredictable. Both warned that structural change and Knightian uncertainty were the reason. Forty years on, we are still finding the same patterns, with sharper methods, at more institutions, and with new regimes to test them on. What has changed is what we can now say about the structure of the deviations.

The interpretive question the issue is built around is not whether forecasts deviate from outcomes in systematic ways. That has been settled empirically since Lovell and confirmed by a huge number of empirical studies. The question is what the systematic deviations mean.

The dominant accounts in the literature — information frictions (sticky information, noisy information, rational inattention) and behavioral biases (diagnostic expectations, extrapolation, overconfidence) — preserve the assumption that the data-generating process is stable and formalize the deviation from full-information REH with constant parameters (a loop I traced in detail in Fixed Models in a Changing World). That implies that the systematic components in forecast errors should be the same over time.

The Riksbank record, the Stillwagon evidence, and the Granziera state-dependence finding together suggest a different reading. The systematic components in forecast errors are themselves regime-shifting, in a way that no constant friction and no constant bias can account for. They are what one would expect if the underlying process undergoes unforeseeable change — and if rational expectations have to be reformulated to accommodate such change rather than abstracting from it.

Paper #1: The Riksbank’s Account of Monetary Policy 2025

Document: Account of Monetary Policy 2025, Sveriges Riksbank, published March 3, 2026. The Riksbank’s annual statutory ex-post accountability report to the Riksdag Committee on Finance. Link.

The Account is the Riksbank’s annual public reckoning with its own monetary policy — a statutory ex-post evaluation tied to a parliamentary hearing and complemented by an every-five-years external evaluation (the Ravn–Wilkins report below is the latest). Few central banks evaluate their own forecasting record this systematically: the ECB does so in occasional Economic Bulletin boxes rather than a dedicated report, and the Bank of England began a comparable standalone Forecast Evaluation Report only in January 2026, following the Bernanke review. The 2025 Account covers a year that took the policy rate from 2.50 to 1.75 percent across three cuts.

The first sections of the Account are largely a record of what the Bank has done over the year. The document opens for the first time with a freestanding Executive Board commentary that reaches for Knightian uncertainty language: tariffs “on a scale not seen since the mid-1930s,” “worrying cracks in the rule-based world order that had prevailed since the end of the Second World War,” “longer-term effects ... remain unclear.” The analytical core of the document, for the Dispatch’s purposes, is Section 3.

What the Account shows. The CPIF inflation forecasts the Riksbank produced in 2023 and 2024 sit systematically below the inflation that actually materialized in 2025 (see Figure 25 below). The under-prediction is modest in absolute terms — Swedish inflation in 2025 was within striking distance of the target — but it is one-directional and persists across vintages and across the year. Figure 28 adds a second feature: decomposed by horizon, the errors are signed in the under-prediction direction at every horizon and grow as the horizon lengthens. The Account treats the horizon growth as the salient finding, noting that the judgment-augmented Riksbank forecast carries more long-horizon bias than the underlying MAJA (the Bank’s flagship DSGE) and BVAR model forecasts, and recommends placing greater weight on the model forecasts at long horizons.

A structural-change reading. The observation that matters most is not the horizon growth the Account emphasizes but the persistence it passes over. Under the standard theoretical assumption that the economy’s structure is fixed — the assumption on which MAJA and the BVAR both rest — forecast errors should be unpredictable: mean-zero, with no systematic sign. A run of one-directional errors that persists across an entire episode is precisely the pattern that assumption rules out. It is, however, exactly what one expects when a model estimated with constant parameters is used in an economy whose parameters have shifted: the constant estimates are, in effect, an average over the distinct regimes in the sample, so in any regime that sits away from that average the model is biased in one direction and stays biased for as long as the regime lasts. The horizon growth then follows as a symptom of the same cause rather than an independent fact. A constant-parameter model mean-reverts to its estimated steady state at its estimated speed — here, toward the 2 percent target. The current regime may differ on either margin: a different anchor, or a slower convergence to it. Since the anchor is ultimately the fixed target, the live margin is the convergence rate — so the forecast is pulled back to target faster than the regime warrants, and the bias grows with the horizon.

The judgment-helps-on-average finding. Section 3.2 concludes that, across all forecasted variables for the studied period, “judgement has, on average, contributed to more accurate forecasts” (page 47) — though not for inflation, where judgment adds to the long-horizon bias rather than reducing it (Figure 28). It is the rare case in which an institution publicly weighs its own judgment against its model, and the result is telling. Judgment is where the forecasters’ reading of the current regime enters: how the situation differs from the past, what has changed recently, where the conjuncture seems to be heading. The models, by the fixed-structure assumption, cannot represent any of this — they carry forward the average parameters of the estimation sample. So the fact that adding judgment improves the forecasts on average is itself a symptom that the models are missing the regime change.

The forecast errors in two figures

Figure 25 records the systematic component over the year; Figure 28 decomposes the errors by horizon and compares the judgment-augmented Riksbank forecast against the underlying model forecasts.

Figure 25: CPIF forecasts produced in 2023 and 2024 against the inflation that materialized in 2025. The forecasts sit systematically below the realized path throughout the year.

Figure 28: CPIF forecast errors by horizon. The errors are signed in the under-prediction direction across all horizons and grow with the horizon. The judgment-augmented Riksbank forecast carries more long-horizon bias than the underlying MAJA (DSGE) and BVAR model forecasts.

Reprinted from Sveriges Riksbank, Account of Monetary Policy 2025.

Why I’m recommending it

• The Account is a detailed public record of a major central bank’s recent forecast performance, with judgment-augmented forecasts compared against MAJA and BVAR forecasts at horizons of 0 to 12 quarters.

• Figures 25, 26, and 28 document systematic, persistent under-prediction of 2025 inflation that the Account narrates as a horizon issue but that I read as a symptom of models that do not account for unforeseeable change.

• The page-47 finding that judgment helps on average is the clearest public statement from inside an institution that the judgment can improve model forecasts.

The takeaway

The 2025 record shows that the same constant-parameter model and the same judgment apparatus that the Riksbank was using before the post-pandemic period continue to produce systematic forecast errors, smaller in magnitude than in 2021–22 but still patterned. The recommendation to weight the model more heavily at long horizons accepts the constant-parameter model as a suitable theoretical framework. The deeper question — whether constant-parameter models are appropriate for forecasting and policy analysis in a changing world — is not asked.

If you only read a few pages

Read Section 3 in full. Figures 25 and 26 are the systematic-component record. Figure 28 is the horizon decomposition. Page 47 carries the “judgment helps on average” finding. The Executive Board commentary in Section 1 is worth a quick read for the Knightian-flavored passages.

Paper #2: Ravn and Wilkins on the Riksbank’s Forecast Performance

Document: Riksbank Evaluation, 2015–2024 by Morten Ravn (University College London) and Carolyn Wilkins (Princeton; former Bank of Canada Senior Deputy Governor; current Bank of England Financial Policy Committee external member). Commissioned by the Finance Committee of the Sveriges Riksdag, delivered January 12, 2026. Link.

The Ravn–Wilkins report is the fifth in the series of independent external evaluations of Swedish monetary policy in a longer-term perspective that the Riksdag Committee on Finance has commissioned since the mid-2000s. At 146 pages, ten chapters, and seventeen recommendations, it is the most comprehensive external evaluation of any major central bank’s monetary policy in 2026. The headline assessment is that the Riksbank acted with determination in difficult circumstances and that targeted reforms are warranted rather than a wholesale redesign of the framework.

Chapter 6, on forecast performance, is the most relevant part for this Dispatch. Figure 22 shows the Riksbank’s CPIF inflation forecasts made through 2021 and 2022 against the realized path: the forecasts sit far below realized inflation throughout the surge, predicting a steady return to the 2 percent target while inflation kept climbing, with under-predictions that were large and sustained over many quarters. And the Riksbank was not alone. The 2021–22 inflation surge was a global phenomenon, and the report’s cross-country comparison shows that forecast errors of this kind — large, one-directional, and persistent — hit virtually every major central bank. This was a period of major structural change, in supply chains, energy and labor markets, and the pricing behavior of firms; it is the most important recent example of a forecast breakdown caused by unforeseeable structural change.

What the evaluation says went wrong. Chapter 6.3.2 is direct about why MAJA — the Riksbank’s flagship DSGE model — broke down. Because MAJA is fitted to historical data, the report notes, “it is aimed at accounting for economic circumstances that have been observed in the past,” and so “is unlikely to properly account for unusual circumstances that have rarely been observed in the data.” The chapter then itemizes the assumptions that failed: linear exchange-rate pass-through, against documented threefold asymmetry across inflation regimes (Linderoth and Meuller 2024 find 17.4 percent pass-through in high-inflation regimes versus 6.9 percent in low-inflation regimes; MAJA is linear); a fixed Calvo price-setting frequency, against the faster pass-through documented for 2021–22 (Klein, Strömberg, and Tysklind 2024); mean-reverting inflation forecasts; and symmetric, linear shock responses around a steady state. The report cites Leeper (2003) on DSGE models missing “structural uncertainty,” and Iversen, Laséen, Lundvall, and Söderström (2016) on Riksbank DSGE underperformance during structural breaks, when judgmental overlays come to dominate the model’s own output.

A comment. The evaluation sees the problem clearly. It does not treat 2021–22 as bad luck — it even diagnoses the policy conduct of those years as “fighting the last war” — and it states plainly that “the priority now is to adapt [the Riksbank’s] tools to an environment in which large shocks and structural breaks may be more common” (page 9). Where I would push further is on what adapting the tools requires. The fixes the report recommends — make pass-through nonlinear, recalibrate Calvo, allow some asymmetry, lean more on scenarios — each relax one assumption at a time, but all of them keep the model’s parameters constant; they make a fixed structure a little richer rather than letting the structure change. That is the same constant-parameter assumption behind the Account’s persistent 2025 misses in Paper #1. Taking unforeseeable change seriously means building it into the formulation of the model itself, and estimating its parameters while continuously testing for structural change rather than assuming them to have been constant without testing this crucial assumption. I set out the policy side of this in the April Dispatch and the forecasting side in Five Principles for Forecasting in a Changing Economy.

There is a further reason to take that deeper move seriously. The report reads 2021–22 as an extraordinary episode. Indeed it was. But “extraordinary” implies a normal to return to, and events since publication cut against that reading. The Iran war and the surging energy prices that followed it arrived only weeks after the evaluation appeared. The world has not settled back into the pre-2021 regime, it has moved into another one. If structural breaks are a recurring feature of the environment rather than rare departures from a stable structure, then opening models and forecasting practice to unforeseeable change is not a contingency plan for the next crisis. It is the normal case.

The forecasts vs. realized inflation in one figure

Figure 22, from Chapter 6, is the 2021–22 counterpart to the Account’s 2025 record.

Figure 22: Riksbank CPIF inflation forecasts made in 2021 and 2022 against the realized path. The forecasts sit far below realized inflation throughout the surge — the models predicted a return to the 2 percent target while inflation kept rising.

Reprinted from Ravn and Wilkins (2026), Riksbank Evaluation 2015–2024.

Why I’m recommending it

• Chapter 6.3.2 is the most detailed external evaluation of a major central bank’s monetary policy in 2026, and it names the specific structural-stability failure mode of its flagship DSGE model: linear pass-through, fixed Calvo, mean-reverting forecasts, symmetric responses.

• The 2021–22 surge is the most important recent example of a forecast breakdown caused by unforeseeable structural change, and the report shows it was no Swedish idiosyncrasy: errors of the same kind — large, one-directional, persistent — hit virtually every major central bank. It is the cleanest available case of constant-parameter forecasting failing systematically when the structure shifts.

• Reading the report alongside the Account gives the closest thing available to a controlled comparison: same model, same institution, same forecasting apparatus, two episodes of systematic under-prediction with very different magnitudes. The pattern is the empirical signature the literature has been chasing for forty years.

The takeaway

Ravn and Wilkins diagnose the forecast breakdown clearly and with independent scholarly authority: they grant that MAJA works in normal times but not in unusual ones, and they call for adapting the Riksbank’s tools to a world of more frequent structural breaks. The fixes they propose stay within the constant-parameter framework — which is why the evidence the report collects supports a deeper reading than the evaluation itself reaches for.

If you only read a few pages

Read Chapter 6 in full. Section 6.2 with Figure 22 carries the 2021–22 forecast record, and the report’s international comparison shows the same breakdown across central banks. Section 6.3.2 explains why MAJA broke down. Section 6.3.3 is the scenarios discussion. Recommendation 3.3 — institutionalize scenarios as a decision device — is in Chapter 10.

Paper #3: Stillwagon on Structural Breaks in Forecast-Error Regressions

Paper: “Professional forecasters do not commit timeless errors: Evidence from structural breaks in tests of over-reaction” by Joshua R. Stillwagon. Economics Letters, vol. 263, article 112939, 2026. Link.

Stillwagon applies Bai-Perron structural break tests to the Coibion–Gorodnichenko (2015; CG) forecast-error regression using the longest-running Survey of Professional Forecasters inflation series — the GDP deflator — over 1969–2016, ending in 2016 to match the sample of Bordalo, Gennaioli, Ma, and Shleifer (2020; BGMS). The CG regression is the workhorse of the modern forecast-error literature. It regresses the consensus or individual forecast error on the prior forecast revision. The rational expectations hypothesis predicts that the coefficient on the revision should be zero when implemented in a constant-parameter model with full information. CG’s full-sample finding is a significantly positive coefficient at the consensus level — interpreted as the canonical evidence for information rigidity. BGMS, at the individual level, find a negative coefficient — interpreted as the canonical evidence for diagnostic-expectations over-reaction. The literature treats both as stable structural features of expectation formation.

The result of allowing for structural breaks reverses both conclusions (Figure 1). At the consensus level, the full-sample finding of under-reaction holds only post-1999. Earlier subsamples show insignificant revision coefficients but significant, time-varying intercepts. At the individual forecaster level, break-adjusted regressions reveal consistently negative revision coefficients across all regimes at the 1 percent significance level — completely reversing the full-sample conclusion that masked over-reaction by averaging across regimes. The estimated breaks align with Federal Reserve leadership changes, suggesting that genuine uncertainty about novel policy regimes drives forecast errors.

The title says exactly what matters: forecast errors are not timeless. The systematic component shifts at regime boundaries. The paper is an academic complement to the Riksbank record. Where the Account and external evaluation document two episodes of under-prediction with very different magnitudes in the same institution’s inflation forecasts, Stillwagon documents the same kind of regime-contingent structure formally across half a century of US survey data.

A comment. The standard readings of these regressions — information rigidity (CG 2015) and diagnostic over-reaction (BGMS 2020) — share a hidden premise: that the data-generating process is constant and the deviation from FIRE can itself be represented with a constant parameter. Stillwagon’s empirical findings refute that premise: the parameters of the forecast-error regressions are not constant. But the deeper point is the one Roman Frydman and I make in our working paper and in the revision we are completing: FIRE implemented in a constant-parameter model cannot represent the expectations of rational market participants, because it abstracts from the unforeseeable change those participants actually face. Once the economy can move to a regime that has not yet been observed, model-consistent expectations no longer imply unpredictable forecast errors. Instead, they imply the systematic, regime-shifting components we and Stillwagon document.

Full-sample vs. break-adjusted estimates in one figure

Figure 1 is the argument in one picture: the single full-sample estimate set against the regime-specific estimates the breaks reveal.

Figure 1: The Coibion–Gorodnichenko revision coefficient, estimated over the full sample versus over the break-adjusted subsamples. The single full-sample number masks regime-specific estimates that differ across periods — the constant coefficient is an average over distinct regimes.

Reprinted from Stillwagon (2026).

Why I’m recommending it

• The paper is a current academic complement to the Riksbank record. Same kind of finding — regime-contingency in the systematic component of forecast errors — produced with a different econometric method, on a different sample, over a much longer horizon.

• The empirical results is a direct empirical refutation of the methodological premise behind both the CG (2015) information-rigidity interpretation and the BGMS (2020) diagnostic-expectations interpretation.

• The empirical results suggest that accounting for survey-based expectations requires acknowledging the importance of unforeseeable structural change.

The takeaway

Forecast errors are not timeless. The systematic component shifts over time. Accounting for survey-based forecast errors requires acknowledging unforeseeable structural change.

If you only read a few pages

Read Section 3 for the consensus-level results and Section 4 for the individual-level results. Figure 1 shows the full-sample versus break-adjusted estimates side by side.

Paper #4: Granziera, Jalasjoki, and Paloviita on State-Dependent ECB Forecast Bias

Paper: “The Bias of the ECB Inflation Projections: A State-Dependent Analysis” by Eleonora Granziera (Norges Bank), Pirkka Jalasjoki (Bank of Finland), and Maritta Paloviita (Bank of Finland). Journal of Forecasting, vol. 44, no. 3, pp. 922–940, 2025. Link.

Granziera, Jalasjoki, and Paloviita test for state-dependent bias in ECB inflation projections using the Odendahl–Rossi–Sekhposyan (2023) threshold regression approach. Linear, full-sample tests find ECB forecasts unbiased on average. The state-dependent tests reveal systematic under-prediction when observed inflation exceeds an estimated threshold of 1.8 percent — close to the ECB’s de facto target — with bias peaking at intermediate horizons (one to four quarters ahead) and ranging from 0.10 to 0.37 percentage points. The bias cannot be explained by errors in conditioning assumptions for interest rates, exchange rates, or oil prices.

The 2021–22 surge is excluded — and the result survives. The baseline analysis deliberately strips the largest forecast errors out of the sample as outliers: the vintages of the Great Financial Crisis, the COVID shock, and the 2021–22 surge are all dropped (see Table A1). The state-dependent under-prediction is therefore not an artifact of the extreme episode — it is identified off ordinary above-target quarters, because the estimated 1.8 percent threshold is low enough that most non-crisis periods sit above it. As the authors put it, the bias “is the result of systematic underprediction by the ECB even at times when inflation does not substantially overshoot the target” (p. 923), and it holds even when the entire post-2019 sample is removed.

This is the piece that completes the picture in this issue. Ravn and Wilkins document the Riksbank under-predicting through the 2021–22 surge; the Account finds the same under-prediction in the much milder 2025 episode; Stillwagon finds the systematic component of US professional forecasters’ errors shifting across half a century; and Granziera, Jalasjoki, and Paloviita find it at the ECB with the surge taken out altogether — in the ordinary above-target quarters that came before and around it. These systematic forecast errors are not a one-off event. They appear across central banks and professional forecasters, and they shift over time — which is what a regime-shifting process produces, and what a fixed bias or a one-off shock does not.

In one quote

The ECB systematically underpredicts inflation when inflation is high, which results in projections converging to target too quickly.

Granziera, Jalasjoki, and Paloviita (2025), p. 937.

Why I’m recommending it

• The paper documents the threshold-conditional form of the finding that the Riksbank record and Stillwagon establish from different angles. Three independent econometric approaches, two central banks and professional forecasters, one phenomenon.

• The paper documents a systematic bias in the ECB forecasts that occurred not only during the 2021-22 inflation surge, and which is invisible in the linear and constant form that averages across the entire sample.

• The Odendahl–Rossi–Sekhposyan threshold regression is one of the central tools the modern forecast-evaluation literature is now using to detect regime-contingent bias that averaged tests miss.

The takeaway

ECB projection bias is not constant. It is concentrated in periods where inflation exceeds 1.8 percent.

If you only read a few pages

Read Section 3 for the threshold regression results and Table 1 for the contrast between linear and state-dependent bias estimates. Figure 1 shows the inflation states and projection distributions. Figure 2 (the bias curve as a function of the inflation state) is the cleanest single visualization of the state-dependent bias. Section 5.2 shows that standard models exhibit the opposite-sign bias.

The Essentials: Zarnowitz (1984) and Lovell (1986) on Testing Expectations with Survey Data

Papers: “Business Cycle Analysis and Expectational Survey Data” by Victor Zarnowitz. NBER Working Paper No. 1378, June 1984. Link. And “Tests of the Rational Expectations Hypothesis” by Michael C. Lovell. American Economic Review, vol. 76, no. 1, pp. 110–124, March 1986. Link.

Forty years ago, testing the rational expectations hypothesis against survey data was a contested move. The objection was methodological: expectations are not observed, so — in Edward Prescott’s words, quoted by Lovell — “surveys cannot be used to test the rational expectations hypothesis.” Zarnowitz and Lovell rejected that view and built the case that survey forecasts are exactly the right instrument for testing theoretical representations of expectations. Lovell held that direct testing of the hypothesis is “an appropriate and worthwhile activity”; Zarnowitz, that “it is not good ‘positive economics’ to dismiss [the survey evidence] on the ground that only theories, not their assumptions, can be tested.” The two papers are complementary in coverage — Zarnowitz examines professional forecasters (the Livingston and ASA-NBER surveys), Lovell the anticipations of firms and households (sales, inventories, prices, wages) alongside government forecasts — so that together they survey the whole field. The skepticism they argued against is long gone: a vast modern literature now does exactly what they advocated (see Coibion, Gorodnichenko, and Kamdar 2018 for a review).

Both papers show how regressions of actuals and survey forecasts can be used to test different theoretical representations of expectations as specific restrictions on the parameters. They were not the first to evaluate expectation representations this way: the framework descends from Mincer and Zarnowitz (1969). What followed was an enormous literature that extended these regressions in every direction and proposed new representations of expectations — sticky information, noisy information, diagnostic expectations — to account for what the tests kept finding.

What the tests kept finding is that the (full-information) rational expectation hypothesis’ prediction of unpredictable forecast errors is rejected by the data, most decisively for inflation. Using the ASA-NBER survey, Zarnowitz finds that roughly 70 percent of individual inflation forecasts fail the unbiasedness test and two-thirds fail the test for serially uncorrelated errors, against about 20 percent for real output. His verdict: “the great majority of inflation predictions since the late 1960s fail the rationality tests and show a strong underestimation bias,” for experts and agents alike. Lovell, reviewing the firm and household evidence, reaches the same conclusion (quoted below). Decades of work since have confirmed it — and yet a great deal of macroeconomic theory still rests on the hypothesis these papers showed the data reject.

Most striking, reading the two papers now, is the explanation they offered and that the literature then forgot. Both located the problem in structural change and genuine uncertainty. Zarnowitz is explicit: the rational expectations hypothesis presumes a stationary environment that agents have had time to learn, so that “unlike in Knight, 1921, or Keynes, 1936, there is no uncertainty here as to what the applicable objective probability disturbances are.” In a world of structural change, he argued, “uncertainty in the sense of Knight and Keynes is prevalent,” and the right notion of rationality is the effective use of “the limited available knowledge and information” in the face of an inherently uncertain future. Lovell, more cautiously, noted that departures from rationality “may be a transient phenomenon arising because economic actors are learning to adapt to a shift in regimes.” The literature that followed kept the rejection but set the diagnosis aside — formalizing the deviation from rationality with fixed parameter inside a stable model, rather than as the rational response to a changing one.

In two quotes

“the weight of empirical evidence is sufficiently strong to compel us to suspend belief in the hypothesis of rational expectations” — Lovell (1986)

“In a nonstationary world with structural changes and a mixture of random and autocorrelated disturbances, uncertainty in the sense of Knight and Keynes is prevalent.” — Zarnowitz (1984)

Why these are the essentials

• They are the foundational empirical case that survey forecasts can and should be used to test theories of expectations — made when much of the profession doubted it could be done at all. The entire modern forecast-error regression literature, including the four new papers in this issue, descends from the argument these two papers won.

• They reached, with the tools of the 1980s, the conclusion that the (full-information) rational expectations hypothesis is rejected by survey forecasts, most decisively for inflation — a finding confirmed by four decades of work since.

• They named the reason — structural change and Knightian uncertainty — and proposed that rationality be understood as the effective use of limited knowledge under genuine uncertainty. That diagnosis was largely forgotten. Recovering it is what this issue is about.

The takeaway

Zarnowitz and Lovell established forty years ago that survey forecasts reject the full-information rational expectations hypothesis, and warned that structural change and Knightian uncertainty were the reason. The literature kept the rejection and forgot the warning, formalizing the deviation with a constant parameter inside a stable model. The four new papers suggest that the deviation is not constant.

If you only read a few pages

In Zarnowitz, read Section 2 (the critique of rational expectations) and Section 5 (the concluding observations), with Table 2 for the rejection rates. In Lovell, read Sections I–II for the framework and the evidence, and Section IV — “Should the Facts Be Allowed to Spoil a Good Story?” — for the assessment.

Closing Remarks

That’s it for the May issue of the Knightian Uncertainty Dispatch.

We have known for forty years that survey forecasts reject the full-information rational expectations hypothesis. Zarnowitz and Lovell established it in the mid-1980s — and both warned that the reason lay in structural change and uncertainty of the kind Knight and Keynes described.

The literature kept the rejection and set the warning aside. It explained the deviations from rationality by adding frictions — sticky information (Mankiw and Reis 2002), noisy information and rational inattention (Sims 2003; Woodford 2003), the forecast-error-regression revival (Coibion and Gorodnichenko 2015) — or behavioral biases — diagnostic expectations (Bordalo, Gennaioli, and Shleifer 2018) and extrapolation. Each of these deviations from REH’s perfect-foresight benchmark under full information is formalized with constant parameters in models that maintain the assumption that the economy’s structure is fixed. The future is assumed to be a probabilistic replica of the past. Frictions or behavioral biases lead to exactly the same systematic components in forecast errors over time.

The four papers in this issue are the evidence that the systematic components in forecast errors are not constant. The Riksbank’s own forecasts under-predict inflation by a margin that shifts with the regime; Stillwagon shows the Coibion–Gorodnichenko coefficient breaks at regime boundaries; Granziera, Jalasjoki, and Paloviita show the ECB’s bias is concentrated in regimes far from target and invisible to averaged tests. The systematic component of forecast errors is itself regime-shifting — which is exactly what a constant friction or a constant bias cannot represent.

Roman Frydman and I read this differently. The empirical evidence from forecast-error regressions does not reject the presumption that market participants are rational. It rejects the assumption that the economy’s structure is fixed — the assumption that the future is merely a probabilistic replica of the past — that REH rests upon. In other words, full-information REH is not an adequate theoretical representation of rational market participants’ expectations, because it abstracts from the unforeseeable change that they actually face.

Once we acknowledge that the economy can shift to a new regime that has not yet been observed, rational participants — using all the information available to them — will still produce forecasts that deviate from outcomes in systematic, regime-shifting ways. The shifting systematic component is not a failure of rationality or a defect of information processing; it is the unavoidable consequence of unforeseeable change. That is the answer to the question this issue opened with — what the systematic deviations mean — and it points to a constructive task: take unforeseeable change seriously by building it into our theoretical models and into how we forecast.

Building unforeseeable change into a model and representing expectations as consistent with such change alters the notion of rational expectations itself. They shift over time, in anticipation of and in response to unforeseeable change in the economy; they deviate from outcomes even when participants have full information; they differ across participants even when those participants share the same information; and they are shaped in part by psychological factors. My Substack series Rational Expectations Under Knightian Uncertainty develops this — how opening a model to unforeseeable change reshapes model-consistent expectations and, with them, what rational expectations are.

Roman Frydman and I presented the framework in two working paper in 2025. The first, Rational Expectations of Inflation Undergoing Unforeseeable Change, shows how to open economic models to unforeseeable change and represent participants expectations as consistent with such change. The second, Unforeseeable Change in Rational Participants’ Inflation Expectations, applied it empirically to forecast-error regressions. We are currently finalizing a substantially revised version of the latter. It uses two complementary break-detection methods — Bai-Perron and indicator saturation with Autometrics — to identify the nonrepetitive parameter shifts in U.S. Survey of Professional Forecasters forecast errors that constant-parameter REH and Markov-switching FIRE both rule out. The patterns this Dispatch documents — at the Riksbank, in the academic literature, and at the ECB — are exactly what that framework predicts.

If you have suggestions for papers I should cover in future issues — especially work that connects structural change, Knightian uncertainty, and real-world forecasting and policymaking — please send them my way.

And if you found this Dispatch useful and want the next issue in your inbox, consider subscribing. It helps the Dispatch reach the people who are interested in developing economic theory, policy analysis, and practical forecasting tools for a changing world characterized by Knightian uncertainty.

Welcome to the fifth issue of the Knightian Uncertainty Dispatch — a monthly curated reading list for anyone thinking about macroeconomics, finance, and forecasting in a world where the future is not just risky, but sometimes genuinely unforeseeable because it may differ from the past.

Each month, I recommend four new papers + one “essential” that help grapple with structural change and uncertainty beyond probabilistic risk — and what those realities imply for economic modeling, forecasting, and policymaking.

The goal is to curate papers that:

1. Deepen our understanding of market outcomes and policy in a world with unforeseeable structural change and Knightian uncertainty.

2. Are useful for how we actually reason and forecast in unstable environments.

3. Connect to one another — so the pieces speak to each other rather than living in isolated silos.

This month’s theme picks up where the March Dispatch and my post “The Quiet Revolution in Central Bank Forecasting” left off. Together, the two pieces documented a remarkable institutional shift unfolding in real time. As the surging oil prices caused by the war in Iran put central banks under pressure, seven major institutions responded with strikingly different uncertainty-communication approaches — and three that had previously published probabilistic forecasts moved to scenario-based communication within six weeks of each other. The Quiet Revolution post traced the cross-bank picture; the March Dispatch curated the primary speeches and policy documents behind it.

This issue asks the deeper question: what does it mean for monetary policy when the economy may be undergoing structural change?

When we acknowledge that the economy may be changing to a new regime that we have not yet observed, three implications follow for monetary policy. They are not new ideas. They are simply what it means to take structural change seriously.

First, neither the central bank nor market participants can know which scenario will unfold. The future regime is, in a real sense, not yet written. We can’t know exactly what it will look like. Even rational and well-informed individuals and institutions will base decisions on expectations that may not align with what will actually happen. This gives rise to large and systematic forecast errors at exactly the moments when forecasts matter most — the moments at which the regime is shifting.

Second, policymakers and market participants need not share a common scenario. Under unforeseeable structural change, there is no single correct model that both parties will converge to. The central bank may be reading the future regime differently than the market; the market may itself be heterogeneous. This is the coordination problem: monetary policy may not have its desired effect, not because the central bank lacks credibility, but because the policymaker’s view of the transmission mechanism differs from the market’s view.

Third, central banks need to consider robustness across multiple scenarios rather than optimization given a single scenario. When neither alignment nor coordination of expectations can be assumed, the right policy is the one that performs reasonably well across the range of plausible regimes — not the one that performs best inside any single scenario. The cost of being wrong about which regime is operative can dominate the cost of being slightly suboptimal under the realized regime.

The four papers in this issue show that senior central bankers are explicitly recognizing each of these implications. A G7 central bank governor names the structural change and the identification problem (Macklem). A chief economist at a major central bank names the coordination problem and makes the case for robustness over optimization (Pill). A Macro Technical Paper from the Bank of England documents how scenarios are now constructed in operational practice based on their flagship DSGE model (Albuquerque et al.). And the April 30 Bank of England Monetary Policy Report — the central document of this issue — published three illustrative scenarios with no designated central projection. The institutional shift the March Dispatch documented has now reached the headline projection vehicle of another major central bank.

What emerges from reading the four papers together is that the policy implications of structural change are visible to senior policymakers and have begun to reshape institutional practice. What is still missing is the formal apparatus that derives these implications from first principles. The constant-parameter REH-DSGE framework that has been the workhorse of monetary economics since the 1970s implicitly assumes both coordination (the central bank and the market share the model) and alignment (their shared expectations correspond to what will actually happen) — assumptions that become visibly untenable under structural change. The essential — Lucas (1976) — is the founding statement of that framework. Reading Lucas alongside the four papers makes the gap between operational practice and theoretical foundations particularly clear.

A note on format. As with the March issue, this Dispatch features primarily central-bank documents rather than research papers. The deliberate institutional shift toward scenario-based communication is moving faster than the academic literature, and the most articulate statements of the policy implications are now coming from inside major central banks. The essential returns to the academic tradition — and to its founding moment.

Paper #1: Macklem on Structural Change at the Bank of Canada

Speech: “Structural Change — Canada at a Crossroads” by Tiff Macklem, Governor of the Bank of Canada. Delivered at the Empire Club of Canada, Toronto, 5 February 2026; reproduced as SUERF Policy Note No. 402 in April 2026. Link.

This speech is the cleanest articulation in 2026 of structural change as the operating environment for monetary policy. Macklem opens by defining structural change explicitly as “the transition between one steady state and the next” — not a deviation around a stable trend, not a larger-than-usual shock, but a change in the underlying economic environment. He identifies a convergence of three structural forces affecting Canada in 2026: a permanent shift in US trade policy, the diffusion of artificial intelligence through the economy, and the slowing of population growth from accelerated immigration in the early 2020s to its slowest pace in decades. None of these is a temporary disturbance. Each is a transition to a new regime, and they are all happening at once. As Macklem put it:

“The impact of these forces on the Canadian economy will not be a temporary cyclical fluctuation. These are deep structural changes that are transforming the economic landscape.”

The most analytically important section is what Macklem calls the cyclical/structural identification problem. He states the problem cleanly: it is hard to know whether a drop in GDP growth is part of a structural trend or a temporary downturn, and getting the call wrong has costs. If the central bank misdiagnoses a structural slowdown as cyclical, it will overstimulate — adding inflationary pressure to an economy that cannot grow faster without it. If it misdiagnoses a cyclical downturn as structural, it will tolerate avoidable slack, leaving people unemployed who could be working. The two errors are symmetric in form but very different in consequence, and the data alone do not tell the central bank which is happening in real time.

This is the first of the three implications stated from inside the policy machinery. The central bank cannot know ex ante which scenario will unfold, and the cost of being wrong shapes the policy choice. Macklem’s response is methodological rather than tactical: the Bank will rely more on scenario analysis to “explore alternative interpretations and reduce the risk of policy errors,” develop richer multi-sector models, and use more granular and regional data to detect structural shifts as they unfold. The framework — the 2 percent inflation target — does not change. How the framework is implemented does.

A comment. The speech is unusually clear about the operational consequences of taking structural change seriously — scenarios, multi-sector models, granular data, the identification problem itself. What it does not do, and could not be expected to do in a policy speech, is ask what theoretical framework would formalize those commitments.

In one quote

“It is hard to know whether a drop in GDP growth is part of a structural trend or a temporary downturn. Getting the call wrong has costs.”

Why I’m recommending it

• A sitting G7 central bank governor explicitly defines structural change as the transition between one steady state and the next, and names the convergence of US protectionism, AI, and demographics as the current rapid-change moment. Read alongside Senior Deputy Governor Rogers’s March 26 remarks, Macklem’s speech gives the Bank of Canada the most coordinated public articulation of any major central bank that the post-pandemic environment is structurally different and that scenarios are the institutional response.

• Macklem names the cyclical/structural identification problem cleanly as the central forecasting and policy challenge.

• Macklem codifies a methodological response in a senior speech — scenario analysis, multi-sector models, granular data — joining the move toward scenarios documented across multiple institutions in the March Dispatch. The framework does not change; how it is implemented does.

The takeaway

Macklem provides the policymaker articulation of the first implication: under structural change, the central bank cannot know which scenario will unfold, and the cost of being wrong shapes the policy choice. The methodological commitment to scenarios follows.

If you only read a few pages

Read the “What is structural change?” section for the explicit definition and the “Monetary policy implications” section for the identification problem.

Paper #2: Pill on Coordination, Robustness, and Eternal Verities

Speech: “Uncertainty, Structural Change and Monetary Policy Strategy” by Huw Pill, Chief Economist of the Bank of England. Maxwell Fry Annual Lecture, Money Macro and Finance Society, University of Birmingham, 8 October 2025. Link.

The Maxwell Fry Lecture is one of the most explicit statements to date by a senior major-bank policymaker that monetary policy must operate under genuine uncertainty about the structure of the economy — not just the realization of shocks within a known structure. The title alone is a Dispatch tagline: three core themes — uncertainty, structural change, and monetary policy strategy — in one phrase, by a sitting MPC member of a major central bank.

Pill’s main message is that “in a world of radical uncertainty and deep structural economic change, more weight should be given to robust eternal verities in running monetary policy, at the expense of pursuing fragile optimising approaches specific to a given set of often ephemeral circumstances.” The argument has two parts. First, when the deep parameters governing the economy — Pill names price-setting behavior, wage-setting behavior, and possibly the formation of expectations themselves — may have shifted, fitted models specific to the previous regime are unreliable guides to policy. Second, when the central bank cannot pin down the natural rate of interest (R*), potential output (y*), or the natural rate of unemployment (u*) with any precision, it must eliminate any uncertainty about its own inflation target (π*) and follow a systematic data-to-decisions mapping that disciplines both private expectations and internal MPC discussions.

This is a generalization of the case for inflation targeting that goes beyond the standard credibility argument. Standard inflation targeting argues that a clear target anchors expectations and reduces inflation volatility. Pill’s argument is that the only thing the central bank can credibly commit to is the target itself, because everything else — the natural rate of interest, the natural rate of unemployment, the slope of the Phillips curve, the response of inflation to energy shocks — is itself uncertain. Conservative central banking, in Rogoff’s 1985 sense and Waller’s 1992 extension, is the right institutional response.

The coordination point. The most analytically distinctive contribution of the speech is Pill’s recognition that under structural change, the central bank and the market need not share a view of the economy. This is where the speech departs most clearly from the standard New Keynesian apparatus. In the standard model with the rational expectations hypothesis (REH), the economist represents both the central bank’s and the market’s expectations by the model’s conditional expectation; any apparent disagreement is a transient information friction. Pill’s setup is different. When the deep parameters of the economy may have shifted, neither party knows the structure for certain, and there is no model to which both could converge. This is the second of the three implications stated from inside the policy machinery. Markets and policymakers need not coordinate on a common scenario, and policy may not have its desired effect because the policymaker’s view of the transmission mechanism differs from the market’s view.

A comment. Pill is unusual among senior central bankers in invoking the term “radical uncertainty” — the term associated with Kay and King (2020), though not formally cited in the speech — and in framing the policy problem in those terms. But the speech operates inside the formal frame of optimal-control monetary policy: the eternal verities Pill anchors on — a clear inflation target, a systematic data-to-decisions mapping, conservative central banking — are themselves derivable inside the standard model under specific assumptions. The speech identifies the question, but the formal apparatus available to it cannot represent the rational response to genuinely structural uncertainty as different from the rational response to risk.

In one quote

“In a world of radical uncertainty and deep structural economic change, more weight should be given to robust eternal verities in running monetary policy, at the expense of pursuing fragile optimising approaches specific to a given set of often ephemeral circumstances.”

Why I’m recommending it

• Pill makes the most explicit statement to date by a senior major-bank policymaker that monetary policy must operate under genuine uncertainty about the structure of the economy — not just the realization of shocks within a known structure.

• Pill articulates the coordination point cleanly: markets and policymakers need not share a view of the economy under structural change. He thereby voices the second of the three implications from inside the Monetary Policy Committee.

• Pill makes the case for robustness over optimization — the third implication. His framework — anchor on the target, systematic data-to-decisions mapping, conservative central banking — chooses robust policy over policy optimized for any single scenario.

The takeaway

Pill provides the policymaker articulation of the second and third implications: under structural change, markets and the central bank need not coordinate on a common scenario, and the right policy is the one that is robust across plausible scenarios rather than optimized for any single one. The eternal verities — most centrally a clear inflation target — are what remain when little else can be known with precision.

If you only read a few pages

Read the main message paragraph for the eternal verities framing and the three-bullet diagnostic of structural features of the UK economy for the substantive content. Footnote 4 on Rogoff (1985) and Waller (1992) gives the conservative-central-banking citation chain.

Paper #3: Albuquerque et al. on the DSGE Machinery Behind Scenarios

Paper: “Decompositions, Forecasts and Scenarios from an Estimated DSGE Model for the UK Economy” by Daniel Albuquerque, Jenny Chan, Derrick Kanngiesser, David Latto, Simon Lloyd, Sumer Singh, and Jan Žáček. Bank of England Macro Technical Paper No. 1, June 2025. Link.

This paper is the canonical 2025 reference for how scenarios are constructed at a major central bank. The Dispatch includes it because anyone who wants to understand what scenario-based central-bank communication is doing analytically has to understand the machinery — and Albuquerque et al. document that machinery in unusual operational detail, including a distinction between two types of scenarios that is not always made explicit elsewhere.

The model is a two-agent New Keynesian DSGE — the successor to COMPASS — with optimizing and rule-of-thumb households, an imported-energy sector, time-varying trends, an expanded shock set, and real adjustment costs. Standard fare for a 2025 central-bank model.

The two scenario classes. Section 6 distinguishes two operationally distinct ways of constructing a scenario.

1. Alternate-shock or conditioning-path scenarios hold the model fixed and impose specific shock or path realizations — for example, a world-trade shock that reduces world-trade growth by one percentage point through 2025. The standard DSGE conditional-forecasting machinery is then applied.

2. Structural scenarios change the model parameters or specification — for example, doubling backward indexation in the price-setting Phillips curve and tripling it in the wage-setting Phillips curve relative to baseline — and re-solve and re-simulate the model.

The May 2025 Monetary Policy Report used both. The “weaker-demand” scenario was built from imposed risk-premium and investment-cost shocks (alternate-shock type). The “higher-persistence” scenario perturbed the Calvo stickiness and indexation parameters of the price- and wage-setting Phillips curves to mimic steeper transmission of cost pressures (structural type), calibrated against the Bernanke and Blanchard (2025) decomposition of post-pandemic inflation.

The paradox. The structural-scenario approach explains how structural change is treated in mainstream central-bank DSGE models — and it shows where the limits of the framework become hardest to ignore. Importantly, these models assume that the parameters are constant. Structural change is represented by shifting a constant parameter of an internally stationary model to a new value. Each scenario is itself a fully specified, constant-parameter alternative model, and policy is computed as the equilibrium inside each. After the Lucas critique, this is the conventional way of doing policy analysis applied to scenario construction.

It has three limitations. The first is internal: shifting the parameters of a constant-parameter model is, strictly speaking, inconsistent with the model itself. Inside each scenario, agents are forward-looking and form expectations as if the parameters were going to remain at their current values forever. The very thing the structural scenario is meant to capture — that the parameters might shift — is not visible to the agents inside the scenario. They never entertain the possibility that a shift might occur in the future.

The second is the more important from a policy perspective. By representing structural change as a fixed alternative model in each scenario, the framework presupposes that the central bank’s and market participants’ expectations correspond to the scenario and that the scenario actually unfolds. The two assumptions the rest of this issue is built around, coordination and alignment, are baked into the construction. Within any single scenario, neither implication #1 (forecast errors due to misalignment) nor implication #2 (the coordination problem between central bank and market) can arise: agents and the policymaker are operating in the same internally consistent world. Robustness across scenarios — implication #3 — is the framework’s attempt to handle the limitation externally, by comparing policy across analyst-chosen scenarios. The limitation, however, is structural rather than external. It is built into the assumption that each scenario is an internally stationary model.

The third is methodological. The model’s parameters are estimated or calibrated on historical data under the assumption that they have been constant over the sample — yet the structural-scenario approach assumes those same parameters may shift in the future. If parameters can shift now, they may also have shifted across the historical sample. In that case, the estimates are weighted averages over distinct regimes rather than estimates of stable structural relationships. This would explain the well-known poor forecasting performance of this class of models. (See “Fixed Models in a Changing World”.)

All three limitations are inherited from applying Lucas’s policy-analysis framework to scenario analysis. The essential for this issue takes the connection up at length.

Forecast performance in one figure

The paper is unusually candid about the limits of this apparatus. Figure 10 makes the alignment failure visible: the model’s inflation forecasts and the realized data part company through the 2021–2024 surge, with inflation under-predicted in real time and errors well beyond the model’s credibility bands. This is implication #1 in concrete form — the kind of episode in which a constant-parameter model cannot capture what actually happens, and which motivated the structural-scenario approach the paper documents.

Why I’m recommending it

• It provides the canonical reference document for what state-of-the-art central-bank DSGE looks like in 2025, and for how scenarios are constructed in operational practice. The paper rewards any reader who has heard senior policymakers talk about scenarios and wants to see the machinery.

• It draws the distinction between alternate-shock scenarios and structural scenarios as the key methodological move. Structural scenarios capture mainstream DSGE’s quiet operational concession to structural change: the analysts perturb parameters that the model treats as constant because internal empirical analysis suggests they may have shifted.

• It acknowledges the limits of the apparatus with unusual candor. Figure 10 shows the model “struggles most to predict abrupt changes.” Figure 12 shows that even with perfect foresight of conditioning paths, observed inflation sits at the upper edge of the 90 percent credibility band.

The takeaway

Albuquerque et al. show that mainstream DSGE has incorporated structural change operationally — through analyst-chosen perturbations of parameters that the model treats as constant.

If you only read a few pages

Read Section 5.1 on forecast performance against data outturns (Figure 10 is the key visual); Section 5.2 on counterfactual perfect-foresight forecasts (Figure 12); Section 6.1 on the distinction between alternate-shock and structural scenarios; Section 6.2 on the May 2025 MPR scenarios; and the Conclusion for the planned extensions.

Paper #4: The April 2026 Bank of England Monetary Policy Report

Document: “Monetary Policy Report — April 2026”, Bank of England, 30 April 2026. Accompanied by the Monetary Policy Summary and Minutes of the meeting ending 29 April 2026. Link.

Published the day before this Dispatch went out, the April 2026 Monetary Policy Report is the most significant change to Bank of England forecast communication since fan charts were introduced in 1996. What is new is structural rather than the appearance of scenarios per se: the Bank has published alternative scenarios alongside its central projection in many previous Reports — most recently the “weaker-demand” and “higher-persistence” scenarios in May 2025, and going back to the smooth-Brexit and no-deal Brexit scenarios in 2018–2019. The novelty in April 2026 is that the Report publishes three illustrative scenarios — labeled A, B, and C — as the projection, with no designated central case. The Bernanke Review is explicitly cited as motivation. The central projection with a fan chart approach, which the Bank pioneered three decades ago and which became the global standard for probabilistic forecast communication, has effectively been replaced for this Report as the principal projection vehicle. The scenarios are explicitly illustrative rather than exhaustive, and no probabilities are assigned to them — individual MPC members express their own weights, but no Committee-level probability distribution stands behind the projection.

The Bank is yet another major Western central bank to take this step in response to the surging oil prices caused by the war in Iran. “The Quiet Revolution in Central Bank Forecasting” traced the cross-bank picture in detail; the Bernanke Review’s 2024 recommendation is being implemented globally, fast.

The scenarios. The three scenarios in the Report differ in how they handle two sources of uncertainty: the path of global energy prices, and the strength of any second-round effects on domestic inflation. Scenarios A and B keep the price- and wage-setting structure unchanged and vary the energy-price path and the size of second-round effects. Scenario C goes further — it adjusts COMPASS so that households’ and firms’ “recent experience of high inflation play a greater role in shaping inflation dynamics.” This is exactly the structural-scenario move Albuquerque et al. document in Paper #3: deep parameters that the model treats as constants are perturbed to represent a regime in which the inflation-generating process may have shifted.

State-contingent policy. Box G of the Report develops the analytical framework for setting policy across the three scenarios. Citing Söderström (2002), the Box states that “when there is uncertainty about the strength of inflation persistence, it may be better to err on the side of setting policy as if inflation persistence will be significant.” This is the third of the three implications — robustness across scenarios — stated explicitly as the Bank’s analytical framework in the Report itself. The Minutes go further: paragraph 11 records that “the appropriate monetary policy response would be state-contingent.” Robustness across scenarios is no longer an individual member preference; it has become Committee language.

A comment. The Committee voted 8 to 1 in favor of holding Bank Rate at 3.75 percent. The dissent is analytically illuminating. Huw Pill voted for a 25-basis-point hike, and in his statement he ties the dissent directly to the staff scenarios:

“Structural change in price and wage-setting, and the impact on inflation expectations of greater attentiveness to, and salience of, energy and food prices, may strengthen second-round effects beyond what is captured in those scenarios.”

Pill is saying the scenarios understate the upside risk because structural change in price- and wage-setting reaches further than the staff calibration captures. The dissent rests on a claim about the inflation-generating process, not a different read of the data inside a fixed model. As the same author who delivered the Maxwell Fry Lecture six months earlier (Paper #2), Pill’s vote is the same diagnosis applied at a real policy decision. It is also implication #2 — the coordination problem — made visible inside the Committee: a senior member of the same body that produced the scenarios reads the structural change underneath them differently. The new format makes that disagreement visible rather than averaging it away.

The three scenarios in one figure

Chart 3.2 of the Report shows the projected paths of CPI inflation and the output gap under each of the three scenarios. The visual makes the institutional move concrete: three distinct trajectories, none privileged as a central case, with policy to be set robust across the range.

In Scenario A, CPI inflation peaks at 3.6 percent in late 2026 and falls below the 2 percent target by the end of 2027. In Scenario B, inflation peaks marginally higher at 3.7 percent and returns to target by 2028. In Scenario C, inflation peaks at 6.2 percent in 2027 Q1 and remains above target throughout the forecast horizon, with wage growth peaking at 4.6 percent. The output gap widens to between –1.5 and –1.7 percent of potential by end-2026 across all three. As the introduction to Section 3.1 puts it, the chart presents “three scenarios, A, B and C, which help illustrate a range of potential outcomes for the UK economy, without any one being designated as a central projection.”

Why I’m recommending it

• The Report delivers the most significant change to Bank of England forecast communication since fan charts were introduced in 1996. Three illustrative scenarios. No designated central projection. No probabilities assigned. The Bank has restructured its principal projection vehicle.

• Box G and Minutes paragraph 11 state robustness across scenarios as the Bank’s analytical framework and as Committee language. Implication #3 has migrated from speeches into the operational decision rule of a major central bank.

• Pill’s dissent locates the disagreement explicitly in the staff scenarios, making implication #2 — the coordination problem — visible inside the Committee, where the new format reveals regime-change disagreement that fan charts would have averaged away.

The takeaway

The April Bank of England Monetary Policy Report is the live institutional implementation of all three implications. Three scenarios with no central projection acknowledges that the central bank cannot identify which scenario will unfold (implication 1). Pill’s dissent makes regime-change disagreement visible inside the Committee (implication 2). State-contingent and robust policy is now the collective decision rule (implication 3). The institutions have done the operational work.

If you only read a few pages

Read Section 3.1 for the scenario framework and the rationale for dropping the central projection; Box G for the analytical framework on robust policy under inflation persistence uncertainty; Minutes paragraph 11 for the line that puts state-contingent policy in collective MPC language; and Pill’s dissent in the Minutes for the structural-change diagnosis at a real policy decision.

The Essential: Lucas (1976) and the Constant-Parameter REH Approach

Paper: “Econometric Policy Evaluation: A Critique” by Robert E. Lucas, Jr. Carnegie-Rochester Conference Series on Public Policy, vol. 1, pp. 19–46, 1976. Link.

Lucas (1976) is the founding statement of the methodological apparatus that the four papers above are working with — sometimes explicitly, sometimes implicitly. Reading the four papers alongside Lucas is the cleanest way to see what scenario-based central-bank communication is doing, why it is the right operational response to the current moment, and where the gap lies between current institutional practice and the theoretical foundations available to support it.

The diagnosis and the solution. Pre-Lucas policy analysis used reduced-form econometric models estimated on past data. Lucas argued that the estimated coefficients are not deep parameters but reduced-form summaries that depend on agents’ expectations of the prevailing policy regime — if policy changes, expectations change, and the coefficients shift with them. The constructive proposal that emerged from the critique was to treat preferences and technology as deep parameters, model expectations explicitly via the rational expectations hypothesis, derive aggregate behavior from optimizing primitives, and compute equilibria under alternative policy rules. This combination is the foundation of mainstream macroeconomics in 2026. Every central-bank DSGE model, including the Bank of England’s COMPASS suite documented in Paper #3, is descended from it.

From Lucas to scenarios. Lucas asked a specific question: how does a shift in a policy parameter affect economic outcomes? He answered by shifting the parameter and computing the new equilibrium. This is an equilibrium concept — it characterizes the new steady state, not the transition. Within a constant-parameter model with the rational expectations hypothesis, the answer is internally consistent and analytically powerful. By construction, the central bank’s and the market’s expectations coincide, and those expectations align with what the model says will happen in the new equilibrium with the new policy rule.

The same machinery is now used to compute scenarios. One or more parameters of a constant-parameter model are shifted, the new equilibrium is computed, and the resulting paths are presented as the scenario. Inside any single scenario, the coordination of central-bank and market expectations and their alignment with outcomes hold by the same construction as in Lucas’s policy-shift exercise.

But there is an important difference between Lucas’s question and the scenario question. Lucas’s policy shift can be announced — and once announced, the question of whether it propagates as the model implies turns into a question of central-bank credibility. A scenario for a shift in deep or structural parameters cannot be announced. Neither the central bank nor the market can know which scenario will unfold, what the future parameter values will be, or whether their expectations will align with what happens. This is the argument the central bankers in the four papers above make from inside the institution. Macklem names the cyclical/structural identification problem. Pill names the coordination problem explicitly. Albuquerque et al. document the operational workaround — perturbing parameters the underlying model treats as constant. The April Bank of England Monetary Policy Report drops the central projection entirely and makes robustness across scenarios the collective decision rule.

The implication. The constant-parameter REH machinery — which works for Lucas’s question of an announceable policy shift — cannot be relied on alone for policy analysis and scenario computation when the economy’s structure may itself change in ways neither the central bank nor the market can foresee. What is missing is the formal apparatus that derives the institutional responses documented in this issue from first principles, applying the spirit of the Lucas critique one layer deeper than Lucas himself did. Closing that gap is the next step of our research program.

In one quote

“Any change in policy will systematically alter the structure of econometric models.”

Why this paper is the essential

• It provides the founding statement of the constant-parameter REH approach to policy analysis — the apparatus behind every central-bank DSGE in 2026.

• At 28 pages including discussion, the original paper is short, dense, and rewards direct reading. The diagnosis, the constructive proposal, and three worked examples are all in compact form. Most economists today inherit Lucas’s argument through textbooks and DSGE practice rather than from the source.

• Reading Lucas alongside the four papers in this issue is the cleanest way to see why scenario-based central-bank communication is the right operational response to the current moment — and where its theoretical foundations stop short.

The takeaway

The constant-parameter REH apparatus is the field’s solution to the problem Lucas posed, and it works for the question Lucas asked. Applied to scenarios for unforeseeable structural change, where neither central bank nor market can know what is coming, it runs into limits that the institutions are now actively confronting.

If you only read a few pages

The original paper is short — 28 pages including discussion. The opening two sections state the diagnosis; Section 3 sets out the constructive proposal; Section 5 (”Some Examples”) works through three concrete cases. Read with the question in mind: what does Lucas’s argument imply when applied not to changes in the policy rule, but to changes in the structural parameters the policy rule presupposes?

Closing Remarks

That’s it for the April issue of the Knightian Uncertainty Dispatch.

Three implications follow from taking structural change seriously in monetary policy. First, central banks and market participants cannot know which scenario will unfold — alignment cannot be presumed. Second, central banks and market participants need not share a common scenario — coordination cannot be presumed. Third, the right policy is robust across plausible scenarios rather than optimized inside a single one. All three implications are now visible in the language and practice of senior central bankers. Macklem names the identification problem. Pill names the coordination problem. Albuquerque et al. document the structural-parameter scenario machinery. The April Bank of England Monetary Policy Report — published the day before this Dispatch — drops the central projection, makes state-contingent policy collective Committee language, and adopts robustness across scenarios as the explicit decision rule.

What emerges from reading the four papers alongside Lucas (1976) is a clear sequence. Lucas’s modeling framework for policy analysis — constant deep parameters plus the rational expectations hypothesis plus equilibrium computation — became the workhorse of monetary economics and gave us the apparatus that runs the Bank of England’s COMPASS, the ECB’s New Area-Wide Model, the Federal Reserve’s FRB-US, and every other major central-bank model in 2026. Inside any single stable regime, the apparatus might be adequate. Across regimes, when the deep parameters themselves may be shifting and when neither coordination nor alignment can be presumed, the apparatus runs into limits that its own practitioners are now openly acknowledging. The Bank of England’s flagship Monetary Policy Report no longer designates a central projection. The Bank of England’s own DSGE technical paper notes that the model “struggles most to predict abrupt changes” and announces planned work on a non-linear version with bounded-rationality parameters. The institutions are operating outside the apparatus they inherited.

Closing the gap is the work that comes next. The four papers in this issue make the question concrete: how do we represent rational expectations when alignment with realized outcomes cannot be presumed? What does coordination look like when there is no shared model? How is robust policy derived rather than asserted as an institutional norm? The institutions have built the operational responses. The formal apparatus that supports them is the open theoretical question, and the question current institutional practice has now run up against. That is the work the research program on Knightian uncertainty is positioned to undertake.

If you have suggestions for papers I should cover in future issues — especially work that connects structural change, Knightian uncertainty, and real-world forecasting and policymaking — please send them my way.

And if you found this Dispatch useful and want the next issue in your inbox, consider subscribing. It helps the Dispatch reach the people who are interested in developing economic theory, policy analysis, and practical forecasting tools for a changing world characterized by Knightian uncertainty.

Welcome to the fourth issue of the Knightian Uncertainty Dispatch — a monthly curated reading list for anyone thinking about macroeconomics, finance, and forecasting in a world where the future is not just risky, but sometimes genuinely unforeseeable because it may differ from the past.

Each month, I recommend four new papers + one “essential” that help grapple with structural change and uncertainty beyond probabilistic risk — and what those realities imply for economic modeling, forecasting, and policymaking.

The goal is to curate papers that:

1. Deepen our understanding of structural change and Knightian uncertainty.

2. Are useful for how we actually reason and forecast in unstable environments.

3. Connect to one another — so the pieces speak to each other rather than living in isolated silos.

This month’s theme is the revolution in central bank forecasting: from fan charts to scenarios.

On 28 February 2026, the United States and Israel launched air strikes against Iran. The war that followed has disrupted shipping through the Strait of Hormuz — through which roughly 20 percent of the world’s oil supply normally passes — and attacks on energy infrastructure, including the Ras Laffan facility in Qatar, have raised the prospect of lasting damage to global oil and gas supply. Oil prices have surged to levels comparable to 2022. The International Energy Agency has called it the largest supply disruption in the history of the global oil market.

This is not just a large shock. It is a potential structural change in global energy supply — and the uncertainty it creates is not the familiar kind that can be captured by widening a confidence interval around a baseline forecast. Nobody knows how long the conflict will last, whether the infrastructure damage is reversible, how energy markets will reorganize, or how the shock will propagate through an economy that still carries the memory of the 2021–23 inflation surge. The uncertainty is about which world we are entering, not just how far we might deviate from the expected path. In other words, we face substantial Knightian uncertainty, not just probabilistic risk.

Central banks recognized this immediately. At the March 18 meeting of the Riksbank’s Executive Board, Deputy Governor Per Jansson made a remarkable admission: “I actually find it difficult to identify a clear main scenario among all the possible development paths ahead.” A sitting central banker, publicly stating that the concept of a baseline forecast — the single most likely path — is itself problematic.

Jansson was not alone. In the same week, the ECB dropped its fan charts entirely, replacing them with alternative scenarios. The Riksbank framed its decision around three scenarios, stressing that its revised baseline was “highly uncertain this time” and that all three scenarios were needed to understand the policy choice. The Bank of England’s Chief Economist gave a speech on robustness — arguing that optimization within a single model is the wrong approach under “radical uncertainty.” And ECB President Lagarde opened a speech with the words: “We find ourselves yet again in a different world, whose contours are not yet clear.”

I wrote three posts in March tracing this shift as it unfolded:

• What Should a Forecast Look Like When the World Might Be Changing? (March 17), written as the oil price surge was reshaping the outlook, argued that when the economy might be undergoing structural change, a single forecast with a fan chart is the wrong format — and proposed a scenario-based alternative with narratives, conditional forecasts, within-scenario bands, and signposts.

• Risk vs. Knightian Uncertainty: Why the Distinction Matters (March 20), published the day after seven central banks announced their decisions, explained the theoretical foundation: the distinction between risk and Knightian uncertainty rests on the economy undergoing nonrepetitive structural change — exactly the kind of change a war-driven restructuring of global energy supply represents.

• The Quiet Revolution in Central Bank Forecasting (March 27) documented how those seven central banks responded to the same oil price surge with strikingly different uncertainty communication approaches — and argued that scenarios are replacing fan charts because they better match the structure of the uncertainty central banks actually face.

This Dispatch provides the primary source reading behind those three posts — the voices of the revolution itself, and the institutional survey that documents it is happening globally.

A note on format. This issue features three speeches and one institutional survey rather than the usual academic research papers. That is deliberate. While academic work on uncertainty, robustness, and scenario design has been developing for some time, the most visible action right now is inside institutions — in speeches, monetary policy reports, and strategy reviews. The speeches below are primary documents of a paradigm shift unfolding in real time. The formal theoretical foundation — connecting structural change, Knightian uncertainty, and scenario-based forecasting — is still catching up with institutional practice. That is both encouraging and, for those of us working on the theory, a source of urgency.

Paper #1: The ECB President on “A Different World”

Speech: “Navigating Energy Shocks: Risks and Policy Responses” by Christine Lagarde, President of the European Central Bank. Delivered at The ECB and Its Watchers Conference, Goethe University Frankfurt, 25 March 2026. Link

This speech was delivered six days after the ECB’s March 19 monetary policy decision — the meeting where the ECB dropped its fan charts from the staff projections, replacing them with three alternative scenarios. In the projections document, the staff stated explicitly that “the standard computation of the fan charts (based on historical projection errors) would not, in the present circumstances, provide a reliable indication of the high uncertainty surrounding the current projections.” Lagarde’s Watchers speech is the strategic rationale for that decision.

The opening is striking: “But we find ourselves yet again in a different world, whose contours are not yet clear. We are facing profound uncertainty about the path of the economy.” This is not hedging — it is the ECB President stating that the uncertainty the ECB faces is about the shape of the future itself, not just the parameters within a known model. The contours of the new regime are not yet observable.

Lagarde then sets out three principles from the ECB’s updated 2025 strategy. First, assess the nature, size, and persistence of the shock before acting. Second, focus on risks alongside the baseline — “because the effects of significant price shocks on inflation can be non-linear, we need to work with scenarios.” Third, respond in a graduated way depending on the shock’s intensity: look through small temporary shocks; make measured adjustments for large but not-too-persistent overshoots; act forcefully against significant and persistent deviations.

The speech draws an extended comparison between the current energy shock and 2022. Lagarde argues the macroeconomic backdrop is more benign today — lower starting inflation, no demand-supply imbalances, neutral policy stance, neutral fiscal stance. But she flags an important asymmetry: “An entire generation has now lived through its first episode of high inflation — and it may not be as slow to react a second time.” This is a regime-dependent behavioral parameters argument: the same shock may produce different outcomes because the lived experience of 2022 has changed how firms and workers respond.

A comment. The most analytically significant passage is one that could easily be overlooked: “We judged that we were moving into a world of more frequent supply shocks — we have faced at least four major ones since 2020 — and structurally higher uncertainty.” This is not a claim about this particular shock. It is a structural change claim about uncertainty itself — the ECB updated its strategy because it concluded the regime of uncertainty has permanently shifted. During the Great Moderation, the economy appeared stable enough that the future looked like the past — probabilistic risk, quantifiable from historical data, was a reasonable approximation. That world is gone. What has replaced it is an environment where energy markets, trade networks, wage-price dynamics, and geopolitical alignments are all changing in ways that cannot be forecast from past patterns. That is the shift from probabilistic risk to Knightian uncertainty — and Lagarde is describing it in institutional language, even if she does not use the term.

In one quote

“But we find ourselves yet again in a different world, whose contours are not yet clear. We are facing profound uncertainty about the path of the economy.”

Why I’m recommending it

• This is the ECB President articulating Knightian uncertainty in plain language — not as a theoretical abstraction, but as the operational reality facing the world’s second-largest central bank.

• The speech explicitly connects the ECB’s 2025 strategy update to “structurally higher uncertainty” — a structural change claim about the nature of uncertainty itself, not just its magnitude.

• The regime-dependent pass-through argument — that the 2022 experience may have changed behavioral response functions — is precisely the kind of insight that constant-parameter models cannot capture.

• Delivered six days after the ECB dropped fan charts, this is the primary document explaining why the most significant change in central bank uncertainty communication in a decade happened.

The takeaway

The ECB President is describing the world the way the research program on Knightian uncertainty argues it should be described: the future may differ from the past, probabilistic tools calibrated on history fail when that happens, and multiple scenarios — not a single baseline forecast with a fan chart — are the appropriate response.

If you only read a few pages

Read the opening paragraphs through “three principles,” then the section on “Managing Uncertainty” where Lagarde explains the role of scenarios and the 2025 strategy update.

Paper #2: The BoE Chief Economist on Why Robustness Beats Optimization

Speech: “Robustness” by Huw Pill, Chief Economist and Executive Director of Monetary Analysis, Bank of England. Delivered at the National Bank of the Republic of North Macedonia and SUERF conference, Skopje, 24 March 2026. Link

If Lagarde provides the strategic vision for why central banks are shifting to scenarios, Pill provides the analytical framework. His speech is the most theoretically rigorous of the central bank pieces in this Dispatch — and also the most direct in naming the intellectual challenge.

Pill opens by describing the standard view of monetary policy as “engineering” — the inflation forecast targeting regime that “marked the apogee of viewing monetary policy as engineering.” He then argues that “experience over recent years has further challenged the belief that the economy and the transmission of monetary policy were simple enough and/or understood well enough to be characterised by a naïve linear modelling framework.”

The speech draws explicitly on Kay and King’s concept of “radical uncertainty” — a term that encompasses “the Knightian uncertainty familiar from economics texts; the Keynesian uncertainty that arises from the strategic interaction among economic actors; and, more simply, the ‘unknown unknowns’ that plague the policy making environment in practice.” This is a Chief Economist of the Bank of England naming three distinct forms of deep uncertainty and arguing that all three are operationally relevant.

Pill then develops a simple but powerful framework using efficient-locus diagrams to compare Bayesian (weighted-probability) and min-max (worst-case) approaches to policy selection across competing scenarios. The key result comes from the non-convex case: when the efficient locus is not convex, small changes in beliefs can produce discontinuous jumps in preferred policy. This makes scenario diversity analytically necessary — not just good practice — because the policy landscape itself has discontinuities that a single model cannot reveal.

Pill connects this directly to the Bernanke Review, noting that for many stakeholders, Bernanke’s guidance can be distilled to one word: “scenarios.” But he goes further than Bernanke in explaining why scenarios are necessary from a robust control perspective.

In the applied section, Pill argues that structural change in price and wage dynamics — not just shock magnitude — is what makes the current environment difficult. He distinguishes two interpretations of why inflation was more persistent than expected after 2022: a cyclical tightness interpretation (which would be reassuring for the current shock) and a structural change interpretation (which would not). His own assessment: “the burden of proof lies on the side of those seeking to deny a role for structural change.”

A comment. The speech’s most remarkable passage appears in footnote 30: Pill worries that relying on “small perturbations of the baseline scenario in a workhorse New Keynesian model estimated on data largely from the Great Moderation period” would leave him “vulnerable to missing more appropriate policy prescriptions that would derive from a fundamental rethink of the relevant macroeconomic dynamics.” This is a Chief Economist of the Bank of England admitting — in a formal speech — that the standard model may be fundamentally wrong, that he knows it, and that he adjusts his policy deliberations accordingly. It is difficult to find a more candid institutional acknowledgment that constant-parameter models are inadequate under structural change.

In one quote

“Given these concerns, I internalise in my policy deliberations the possibility that small perturbations of the baseline scenario (e.g. slightly increasing the slope of the wage PC in a workhorse New Keynesian model estimated on data largely from the ‘Great Moderation’ period) would leave me vulnerable to missing more appropriate policy prescriptions that would derive from a fundamental rethink of the relevant macroeconomic dynamics (e.g. models of price- and wage-setting behaviour harking back to the 1970s and ‘80s, where real income resistance plays a greater role).” — Footnote 30

Why I’m recommending it

• Pill provides the analytical theory that Lagarde and Seim implicitly rely on. The efficient-locus framework shows why non-convexity makes scenario diversity analytically necessary — not just good practice.

• The explicit naming of Knightian uncertainty, Keynesian uncertainty, and unknown unknowns as distinct and operationally relevant is rare from a sitting central bank Chief Economist.

• Footnote 30 may be the most candid admission in the entire central bank archive that the standard New Keynesian model is inadequate for policymaking under structural change.

• The distinction between cyclical tightness and structural change interpretations of inflation persistence frames the current policy dilemma with unusual clarity — and Pill openly states which side he’s on.

The takeaway

Pill provides the formal case for why robustness — not optimization — is the right approach when facing radical uncertainty. The efficient-locus framework is simple enough for practitioners but powerful enough to show why scenarios are analytically necessary. And footnote 30 is worth reading closely: it is rare to see a sitting policymaker acknowledge this openly that “a fundamental rethink of the relevant macroeconomic dynamics” may be needed.

If you only read a few pages

Read pages 4–5 for the radical uncertainty discussion and the link to the Bernanke Review, then Charts 1–3 (pages 8–11) for the efficient-locus framework, then pages 13–14 for the structural change discussion, and finally footnote 30 (page 18).

Paper #3: The Riksbank Deputy Governor on How Scenarios Actually Work

Speech: “The Role of Alternative Scenarios in Monetary Policy Communication” by Anna Seim, Deputy Governor of Sveriges Riksbank. Delivered at the 2025 ECB Forum on Central Banking, Sintra, Portugal, 2 July 2025. Link

The Riksbank has been publishing alternative scenarios alongside its policy rate path since 2007 — longer than any other central bank. If Lagarde provides the strategic vision and Pill provides the analytical framework, Seim provides something equally important: the operational implementation. This speech is the most detailed insider account of how scenario-based monetary policy communication actually works in practice.

Seim draws the contrast sharply. Fan charts “visualize uncertainty based on historical forecast errors” — they assume the future will look like the past. Scenarios, by contrast, are “coherent macroeconomic narratives about specific developments” that illustrate how the economy and policy would evolve under different structural stories.

The distinction between the two is not just presentational — it is conceptual. Fan charts communicate uncertainty about magnitude within a given model. Scenarios communicate uncertainty about which model — which structural story — applies. When the economy may be undergoing structural change, the second kind of uncertainty dominates, and fan charts are the wrong tool.

The speech’s most compelling exhibit is the December 2024 tariff scenario. In that round, the Riksbank had published a scenario exploring what would happen if U.S. tariffs pushed up Swedish import prices. When the inflation data came in, the outcome matched the scenario’s prediction — but for entirely different reasons. The actual inflation came from temporary food-price shocks, not tariff pass-through. The same inflation outcome, arising from a fundamentally different structural story, required the opposite policy response. The tariff scenario called for tighter policy; the food-price reality did not.

This example is devastating for the fan chart paradigm. A fan chart that showed wider bands around the central forecast would have captured the magnitude of the deviation. But it would have told the policymaker nothing about why the deviation occurred — and therefore nothing about whether to tighten or ease. The scenario, by contrast, provided the narrative structure that made the deviation interpretable.

A comment. What makes Seim’s speech particularly valuable is that she also discusses the challenges of scenario design — questions that most advocates of scenarios gloss over. How many scenarios? How extreme should they be? How detailed? Should they be anchored in specific risks or explore broader structural possibilities? And how should past scenarios be revisited to build credibility? These are practical questions that any institution transitioning from fan charts to scenarios will face, and Seim’s answers are grounded in the Riksbank’s nearly two decades of experience.

In one quote

“In times of radical uncertainty, patterns can change.”

Why I’m recommending it

• The most detailed first-person practitioner account of how scenario-based communication works in practice — from the institution that has been doing it longest.

• The December 2024 tariff example is a definitive illustration of why narrative matters: same inflation outcome, different structural drivers, opposite policy response. This is the example to use when explaining why fan charts are insufficient.

• Seim explicitly uses the phrase “radical uncertainty” and acknowledges that standard econometric models face limitations — an unusually candid admission from a sitting Deputy Governor.

• The discussion of practical challenges — scenario design, calibration, ex-post evaluation — provides operational guidance that the other speeches lack.

The takeaway

Scenarios are not just communication devices — they are analytical tools for regime identification. The December 2024 example shows that distinguishing between structurally different worlds is essential for policy, and that fan charts — which communicate magnitude without narrative — cannot do this.

If you only read a few pages

Read the December 2024 tariff scenario example for the most concrete illustration of why scenarios matter, then the section comparing scenarios and fan charts for the conceptual distinction, and the section on challenges when working with scenarios for practical guidance.

Paper #4: The Global Evidence — 25 Central Banks Shifting

Paper: “Evolving approaches to monetary policy communication in the face of uncertainty: fan charts, scenarios and guidance” by Sarah Bell, Matthieu Chavaz, Boris Hofmann, Daniel Rees, and Matthias Rottner. BIS Quarterly Review, March 2026. Link

The three speeches above describe the shift from a single baseline forecast with fan charts to multiple scenarios at three institutions — the ECB, the Bank of England, and the Riksbank. But is this shift systematic or anecdotal? Bell et al. provide the answer: it is global, it is accelerating, and it is documented across 25 central banks from 2006 to 2025.

The paper proposes a useful taxonomy. General uncertainty — the inherent unpredictability of the future — has traditionally been communicated through fan charts, which provide a graphical representation of confidence intervals estimated from past forecast errors. Specific uncertainty — stemming from identifiable risks or developments — is better communicated through alternative scenarios that describe how the economy and policy would evolve under different conditions. The shift the paper documents is from the first to the second.

The numbers are clear. Scenario analysis has roughly tripled across the 25 surveyed central banks since 2006 — from approximately 15% to 45% of central banks using scenarios to communicate economic outlook uncertainty. Fan chart usage has declined modestly. Qualitative risk discussions have become near-universal. And increasingly, central banks are embedding policy rate projections within their scenarios — providing scenario-contingent forward guidance rather than unconditional rate paths.

The paper also documents the practical failure that accelerated the shift. The post-Covid inflation surge “revealed the drawbacks of descriptive guidance” — fixed forward guidance was widely interpreted as unconditional commitment, tying central banks’ hands precisely when flexibility was most needed. The shift to scenarios is, in part, an institutional response to this failure.

A comment. The paper’s taxonomy of general vs. specific uncertainty maps, imperfectly but usefully, onto the distinction between probabilistic risk and Knightian uncertainty. Fan charts address general uncertainty by quantifying the range of outcomes within a given model. Scenarios address specific uncertainty by exploring how different structural stories lead to different outcomes and different policy responses. The paper cites Knight (1921) explicitly — acknowledging the distinction between “measurable uncertainty (risk) and unmeasurable (true) uncertainty.” The shift it documents is, in effect, central banks moving from tools designed for measurable uncertainty to tools designed for the unmeasurable kind.

The shift from fan charts to scenarios in one figure

Chart 2 shows the evolution of three approaches to communicating economic outlook uncertainty across 25 central banks from 2006 to 2025. Fan chart usage has declined modestly from about 70% to 65%. Qualitative risk discussions have risen from about 50% to 60%. But the striking trend is scenario analysis: it has roughly tripled, from approximately 15% to 45% of central banks. The institutional shift Lagarde, Pill, and Seim describe at their own institutions is not anecdotal — it is a global phenomenon.

Why I’m recommending it

• It provides the cross-bank evidence that the shift described by Lagarde, Pill, and Seim is not three institutions making independent choices — it is a systemic response across 25 central banks.

• The general vs. specific uncertainty taxonomy gives the Dispatch audience a practical vocabulary for the distinction between probabilistic risk (fan charts) and structural uncertainty (scenarios).

• The paper cites Knight (1921) and explicitly distinguishes measurable risk from unmeasurable uncertainty — connecting the institutional shift to the intellectual tradition this Dispatch is built around.

• It documents the failure of fixed forward guidance during the post-COVID inflation surge — an institutional lesson that motivated the shift toward scenarios as a communication tool.

The takeaway

The shift from fan charts to scenarios is not three central banks making independent choices. It is a systemic institutional response to the failure of probabilistic uncertainty tools during structural change. When nearly half the world’s central banks use scenario analysis — triple the share from two decades ago — that is a revolution, not a trend.

If you only read a few pages

Read the taxonomy section on general vs. specific uncertainty, look at Chart 2 for the trend data, and then read the section on scenario analysis for how central banks are using scenarios in practice.

The Essential: The Bernanke Reviews — The Catalyst for the Revolution

Report: “Forecasting for monetary policy making and communication at the Bank of England: a review” by Ben S. Bernanke. Bank of England, Independent Review, April 2024. Link

and

Paper: “Improving Fed communications: A proposal” by Ben S. Bernanke. Hutchins Center Working Paper #102, The Brookings Institution, 2025. Link

Every revolution has a catalyst. For the shift from fan charts to scenarios in central bank forecasting, it is Ben Bernanke’s 2024 review of the Bank of England’s forecasting framework.

The review was commissioned after the BoE’s forecasting failures during the post-Covid inflation surge. But Bernanke’s diagnosis went far beyond “the models were wrong.” He identified a systemic problem with how the BoE — and, by implication, other central banks — communicate uncertainty.

The most consequential recommendation is also the bluntest. Recommendation 11: eliminate fan charts. Bernanke writes:

“Despite their distinguished history, the fan charts as published in the MPR have weak conceptual foundations, convey little useful information over and above what could be communicated in other, more direct ways, and receive little attention from the public. They should be eliminated.”

Fan charts, Bernanke argues, create an illusion of precision about unknowable futures while obscuring the true driver of policy: the central bank’s reaction function — how it would respond if the economy evolved differently from the baseline. The alternative is scenarios. Recommendation 8: publish alternative scenarios alongside the central forecast. Scenarios allow the public to understand the rationale for the policy choice, including risk management considerations, and to see how policy would adapt to different economic conditions.

A year later, Bernanke applied the same diagnostic to the Federal Reserve. The 2025 Brookings paper argues that the Fed’s Summary of Economic Projections and dot plot suffer from analogous problems. The individual projections “provide at best limited insight into why the implied outlook takes the shape it does” and, more importantly, they focus attention on “what each participant sees as the most likely future path for the economy and policy, unhelpfully downplaying the large role of forecast uncertainty in policymaking.”

Bernanke’s proposed remedy is the same: a transparent baseline forecast complemented by alternative scenarios, enabling “quantitative analyses of uncertainty and risks to the outlook” and refocusing communication on the reaction function. He puts it with characteristic directness: “Policy strategies may often be best communicated in terms of policymakers’ reaction function — if this happens, we will do this; if that happens, we will do that — rather than in terms of modal expectations for the economy and policy.”

In one quote

“Despite their distinguished history, the fan charts as published in the MPR have weak conceptual foundations, convey little useful information over and above what could be communicated in other, more direct ways, and receive little attention from the public. They should be eliminated.”

Why these two papers together

One mind applied the same diagnostic to two continents. The 2024 review identified the problem at the Bank of England — fan charts with “weak conceptual foundations,” a forecasting process focused on the single most likely path, and insufficient attention to the reaction function. The 2025 proposal showed the problem is equally present at the Federal Reserve, where the dot plot focuses attention on modal expectations while “unhelpfully downplaying” the role of uncertainty.

Every institutional reform documented in this Dispatch traces back to these reviews. The ECB dropped fan charts (Lagarde). The BoE’s Chief Economist built an analytical framework for robustness (Pill — who explicitly credits the Bernanke Review). The Riksbank refined its scenario practice (Seim). And the BIS documented the global shift across 25 central banks (Bell et al.). Bernanke did not create the movement — the failures of 2021–23 did that — but he crystallized the diagnosis and pointed the way forward.

Reading the two reviews together reveals something important: the same logic applies regardless of institutional specifics. Whether the problem manifests as fan charts (BoE), dot plots (Fed), or staff projections without scenarios (ECB before 2022), the underlying issue is identical. Central banks that communicate primarily through a single baseline forecast, supplemented by probabilistic uncertainty bands, are using tools designed for a world of quantifiable risk. When the economy undergoes structural change — when the relevant uncertainty is about which world we inhabit, not just how far from the baseline we might end up — those tools fail. Scenarios are the institutional response to that failure.

If you only read a few pages

For the 2024 BoE review: read the Executive Summary and Recommendations 7, 8, and 11 for the diagnosis and the key proposals. For the 2025 Fed paper: read pages 12–16 on the reaction function and the proposal for an Economic Review with alternative scenarios.

Closing Remarks

That’s it for the March issue of the Knightian Uncertainty Dispatch.

From Bernanke’s diagnosis in 2024 to Lagarde’s vision, Pill’s analytical framework, Seim’s operational implementation, and Bell’s global documentation — all in under two years. What emerges from reading these five pieces together is not a gradual evolution but a revolution: a fundamental rethinking of how central banks communicate the uncertainty surrounding their forecasts and policy decisions.

The common thread is the recognition that fan charts — probabilistic uncertainty bands calibrated on historical forecast errors — are the wrong tool when the economy may be undergoing structural change. They quantify a kind of uncertainty (how far from the baseline?) while remaining silent about the kind that matters most for policy (which world are we in?). Scenarios address that deeper question by providing narrative-driven explorations of structurally different futures, each with its own policy implications.

What is still missing is the formal theoretical foundation. The institutions are ahead of the theory — they are already doing what the research program on Knightian uncertainty argues for, but without the formal apparatus to explain why scenarios are the right response to structural change. Central bankers like Pill cite Kay and King on “radical uncertainty,” and Lagarde speaks of “a different world, whose contours are not yet clear.” But the connection between structural change, the failure of probabilistic tools, and the superiority of scenario-based communication has not yet been formalized. That is where my research agenda lies.

If you have suggestions for papers I should cover in future issues — especially work that connects structural change, Knightian uncertainty, and real-world forecasting and policymaking — please send them my way.

And if you found this Dispatch useful and want the next issue in your inbox, consider subscribing. It helps the Dispatch reach the people who are interested in developing economic theory and practical forecasting tools for a changing world characterized by Knightian uncertainty.

Welcome to the third issue of the Knightian Uncertainty Dispatch—a monthly curated reading list for anyone thinking about macroeconomics, finance, and forecasting in a world where the future is not just risky, but sometimes genuinely unforeseeable because it may differ from the past.

Each month, I recommend four new papers + one “essential” that help grapple with structural change and uncertainty beyond probabilistic risk—and what those realities imply for economic modeling, forecasting, and policymaking.

The goal is to curate papers that:

1. Deepen our understanding of structural change and Knightian uncertainty.

2. Are useful for how we actually reason and forecast in unstable environments.

3. Connect to one another—so the pieces speak to each other rather than living in isolated silos.

This month’s theme is central banks in uncharted territory: how major monetary policy institutions are rethinking their frameworks when the future may not look like the past.

Something notable is happening across the world’s central banks. After the forecasting failures of 2021–23, institutions are not just patching their models—they are asking harder questions about the kind of uncertainty they face. Is it uncertainty within a known probability distribution, where wider confidence bands suffice? Or is it something deeper—uncertainty about the structure itself, where the distribution may have changed in ways that cannot be quantified from past data?

The four pieces this month map this institutional reckoning:

• Bauer, Berge, Fiori, Loria, and Zhong lay out the Federal Reserve’s new taxonomy of uncertainty for the 2025 framework review—distinguishing state, structural, and expectations uncertainty—and document that several central banks have abandoned fan charts because they fail during structural breaks.

• Cateau, Coletti, and Portelance describe how the Bank of Canada is shifting from precision engineering to risk management, explicitly invoking “Knightian or radical uncertainty” and proposing “thick-line macro”—one of the few central bank publications to use that language.

• Amaral, Ehlers, Shim, and Tombini survey 12 central banks on how they actually handle uncertainty in practice—and reveal a striking gap: most focus on “known unknowns” with quantifiable likelihoods, rarely addressing the possibility that the future may be genuinely unforeseeable.

• Bailey, the Bank of England Governor, provides the structural backdrop: five qualitative headwinds—from supply shocks to deglobalization to AI—that are reshaping the economic landscape in ways that existing macroeconomic frameworks are, in his words, “less well equipped” to handle.

Together, these pieces tell a story about an institutional conversation that is further along than many academics realize—and yet still hasn’t fully confronted the implications of taking structural change seriously.

Paper #1: The Fed’s Taxonomy of Uncertainty

Paper: “Accounting for Uncertainty and Risks in Monetary Policy” by Michael D. Bauer, Travis J. Berge, Giuseppe Fiori, Francesca Loria, and Molin Zhong. FEDS 2025-073 / SF Fed Working Paper 2025-19. Published 2025.

This paper was prepared for the Federal Reserve’s 2025 monetary policy framework review. Apart from the Bank of England staff paper by Haberis et al. (2025), discussed in the December Dispatch, it offers the most systematic treatment I’ve seen of the different kinds of uncertainty that central bankers face—and it does not shy away from acknowledging the limits of quantification.

At the heart of the paper is a three-part taxonomy: state uncertainty (where are we now?), structural uncertainty (have the relationships changed?), and expectations uncertainty (are agents forming expectations about a stable or shifting world?). The first can, in principle, be resolved with more data. The second and third cannot—not fully—because they involve the possibility that the economy’s structure has changed in ways that historical data alone cannot reveal.

That distinction matters. A footnote early in the paper explicitly acknowledges Knightian uncertainty as “unknown unknowns”—a rare concession in a Federal Reserve publication. And the paper documents a practical consequence: fan charts, which assume a known probability distribution around the central forecast, fail during structural breaks. Several central banks have abandoned them.

A comment. The deeper problem with fan charts is not just that they fail during structural breaks—it is what they are taken to represent. A fan chart is typically interpreted as illustrating the uncertainty around a central forecast. But it only captures the probabilistic uncertainty: the part that can be quantified, conditional on the model. Once we recognize that the economy undergoes structural change, there is an additional layer of uncertainty—Knightian uncertainty—about the future structure, about the model itself. That uncertainty cannot be quantified ex ante; it is ultimately subjective, even when informed by careful analysis of past data. Fan charts are silent about it. They show the uncertainty we can measure and say nothing about the uncertainty we cannot.

Why I’m recommending it

• The three-part taxonomy—state, structural, expectations—maps directly onto the concerns that motivate this Dispatch. It gives practitioners a shared vocabulary for distinguishing “we need more data” from “the world has changed.”

• It was written for the FOMC as part of the 2025 framework review. This is not an academic exercise—it is an attempt to shape how the Federal Reserve thinks about uncertainty in its own policymaking.

• The finding that fan charts fail during structural breaks is important. Fan charts embed the assumption that the future is drawn from a known distribution estimated from the past. When that assumption breaks down, they don’t just become imprecise—they become actively misleading, because the actual uncertainty around a forecast has more layers than probabilistic risk alone.

How central banks communicate uncertainty in one table

Table 2 provides a rare cross-country overview of how major central banks communicate uncertainty. The most striking column is “Fan Charts”: five of the eight central banks listed have either dropped them or never adopted them. Only the Bank of England, the ECB (with limited use), and the Federal Reserve still publish them—and even the Fed’s version is a symmetric, forecast-error-based confidence band with a five-year lag before scenarios become available.

The takeaway

The paper’s most important contribution is not the taxonomy itself but what it implies: different kinds of uncertainty require different policy responses. State uncertainty calls for patience, data gathering, and careful statistical analysis. Structural uncertainty calls for robustness, model diversity, and scenario analysis. Expectations uncertainty calls for clear communication—but also humility about what communication can achieve when agents themselves face Knightian uncertainty.

The paper also documents that the profession’s standard tool for communicating uncertainty—the fan chart—fails precisely when it is needed most. That is a significant institutional admission.

If you only read a few pages

Read the Introduction and Section 2.1 for the taxonomy, then go to Table 2 for the cross-country comparison of how central banks communicate uncertainty. If you have more time, Section 3 discusses how policymakers can account for uncertainty in their decisions.

Paper #2: The Bank of Canada’s Turn Toward “Thick-Line Macro”

Paper: “From models to communications: strengthening risk management in monetary policy at the Bank of Canada” by Gino Cateau, Don Coletti, and Annie Portelance. Chapter in BIS Papers No. 163, pp. 51–58. Published 2025.

This short chapter—only eight pages—may be the most direct engagement with Knightian uncertainty I have seen from inside a major central bank. The Bank of Canada describes how it is shifting from “precision engineering” to “risk management” in monetary policy, and it uses the phrase “Knightian or radical uncertainty” by name to explain why.

Their diagnosis is sharp. The base case forecast creates a “false sense of precision.” Models should be treated as “only one perspective among many.” Risk analysis should extend beyond what the baseline model suggests. These are not cautious hedges in an academic paper—they are operational principles that the Bank of Canada is implementing.

The concrete proposals follow from the diagnosis: “thick-line macro” (communicating forecasts as wide bands rather than precise paths), model suites with different structural assumptions, and shifting from mean to mode for the baseline forecast. Each of these acknowledges that the single-model, single-baseline paradigm is inadequate when the economy undergoes structural change.

A comment. What strikes me about this paper is the explicit language. Many central banks practice some version of model humility—running multiple scenarios, widening confidence intervals, hedging their forecasts. But most of this is framed as dealing with “higher uncertainty”—meaning more risk within a known framework. The Bank of Canada goes further: they invoke Knightian uncertainty by name, acknowledging that the kind of uncertainty may be different, not just its magnitude. That is a meaningful distinction, and it is rare in institutional publications.

Why I’m recommending it

• It is one of the very few central bank publications to explicitly invoke “Knightian or radical uncertainty.” That language matters—it signals a conceptual shift, not just a recalibration.

• The three reform priorities—challenging the false precision of the base case, practicing model humility, broadening risk analysis—read like a practical checklist for any institution grappling with unforeseeable change.

• “Thick-line macro” is a memorable and implementable idea: present forecasts as ranges rather than point estimates, making uncertainty visible rather than hidden behind a single number.

The takeaway

The Bank of Canada is making a bet: that acknowledging the limits of what models can tell us—openly and institutionally—leads to better policy than pretending those limits don’t exist. “Thick-line macro” is not about being less rigorous. It is about being honest that a single number with a symmetric fan chart may be the wrong way to communicate a future that is genuinely uncertain.

If you only read a few pages

At eight pages, read the whole thing. It is concise and clearly written. If you must prioritize, focus on the section describing the three reform priorities and the discussion of “thick-line macro.”

Paper #3: What 12 Central Banks Actually Do When Uncertainty Is High

Paper: “Monetary policy decision-making and communication under high uncertainty: insights from a survey of central banks in the Americas and beyond” by Eduardo Amaral, Torsten Ehlers, Ilhyock Shim, and Alexandre Tombini. Chapter in BIS Papers No. 163, pp. 7–30. Published 2025.

If the Bauer et al. paper gives us a taxonomy of uncertainty and the Cateau et al. paper shows one bank’s response, this BIS survey provides the field-level view: what do central banks actually do when uncertainty is high?

The answer, based on a survey of 12 central banks across the Americas and beyond, is revealing. Most institutions responded to the post-2020 period by reducing forward guidance, moving toward gradualism, and expanding scenario analysis. These are sensible adaptations. But the survey also reveals a striking gap: central banks overwhelmingly focus on “known unknowns”—risks that can be assigned probabilities and bounded by historical experience—and rarely address the possibility that the future may involve “unknown unknowns” that fall outside any probability distribution estimated from past data.

In other words, most central banks are becoming more sophisticated about probabilistic uncertainty without fully engaging with the deeper challenge that the Dispatch is built around: what do you do when the economy may undergo structural changes that you cannot foresee or quantify?

Why I’m recommending it

• It provides rare cross-country survey evidence on how 12 central banks handle uncertainty in practice—not in theory, but in their actual decision-making and communication.

• It documents concrete institutional adaptations: reduced forward guidance, more gradualist policy moves, expanded scenario analysis, alternative communication tools. These are useful benchmarks for practitioners.

• Most importantly, it exposes the gap between institutional practice and the conceptual challenge: institutions disagree on the right approach—fan charts, scenarios, qualitative assessments all coexist—yet most are getting better at quantifying “known unknowns” while still largely avoiding the harder question of what to do about “unknown unknowns.”

The communication toolkit

Graph 6 shows what tools central banks actually use to communicate uncertainty in their reports. Fan charts lead, followed by scenario analysis and conditional forecasts. Qualitative assessments, macro-at-risk models, and objective probability distributions trail behind. The pattern is clear: the most widely used tools are those that quantify probabilistic uncertainty—precisely the “known unknowns.” Tools that might help communicate deeper uncertainty remain on the periphery.

Links to other papers in this Dispatch

This paper is published in the same BIS volume (No. 163) as the Cateau et al. chapter and provides its institutional backdrop. Reading them together is instructive: Cateau et al. represents the frontier of how far a central bank has gone in acknowledging Knightian uncertainty, while Amaral et al. shows that most central banks have not gone nearly as far. The gap between the Bank of Canada’s explicit invocation of radical uncertainty and the field’s focus on quantifiable risk is one of the most interesting tensions in this issue.

The takeaway

Central banks have made real progress in dealing with uncertainty since 2020. They communicate more cautiously, move more gradually, and rely more on scenarios. But most of these adaptations assume that uncertainty can be quantified—wider confidence bands, alternative scenarios with assigned probabilities, data-dependent forward guidance. The harder question—what to do when the probability distribution itself may have changed—remains largely unanswered at the institutional level.

If you only read a few pages

Read the Introduction for the survey design, then skip to the sections on scenario analysis and communication tools for the most actionable institutional comparisons. Graph 6 shows at a glance what visualisation tools central banks use to communicate uncertainty.

Paper #4: The Structural Backdrop—Why All of This Is Urgent

Speech: “The World Today” by Andrew Bailey, Governor of the Bank of England and Chair of the Financial Stability Board. Delivered at the AlUla Conference for Emerging Market Economies, IMF/Saudi Ministry of Finance, 8 February 2026.

This is not a research paper—it is a speech. But it is a speech by the Governor of the Bank of England that frames the current economic moment as one of qualitative structural change, and it provides the backdrop that makes the other three papers in this Dispatch urgent.

Bailey organizes his remarks around five structural headwinds that are reshaping the global economy:

1. Shocks: The nature and frequency of supply-side shocks have changed.

2. Growth potential: Potential output growth is declining in advanced economies.

3. Demographics: Aging populations are altering labor markets and savings patterns.

4. Trade: Deglobalization and trade fragmentation are reversing decades of integration.

5. Finance: The financial system is undergoing structural shifts.

What makes the speech notable is Bailey’s framing. He draws on Schumpeter’s concept of “discrete rushes” of innovation—the idea that technological change does not arrive smoothly but in bursts that create qualitatively new economic configurations. Applied to AI and robotics, this means the current moment is not just “more uncertainty” but a different kind of economic landscape than the one our models were built for.

The admission is direct: “our macroeconomic frameworks are less well equipped” for supply-side shocks of this kind.

A comment. Bailey’s speech is useful not for any single analytical insight—the five headwinds are well known individually—but for what it represents: a senior policymaker organizing the current moment around qualitative structural change, not cyclical fluctuation. The Schumpeterian framing is particularly apt. Schumpeter’s “creative destruction” is precisely the kind of non-repetitive structural change that Knightian uncertainty is about—change that creates genuinely new configurations rather than drawing from a known distribution of past states. When the Governor of the Bank of England frames the world this way, it validates the premise that forecasting and policymaking need to be built around structural change, not just wider error bands.

Why I’m recommending it

• Bailey structures the entire speech around qualitative structural shifts—not cyclical fluctuation, not just “bigger shocks”—which is exactly the distinction this Dispatch insists on.

• The direct admission that existing macroeconomic frameworks are “less well equipped” for the current environment is significant coming from a sitting central bank Governor and FSB Chair.

• The Schumpeterian “discrete rushes” framing provides accessible language for discussing structural change that is qualitatively different across periods—useful for the Dispatch audience.

The takeaway

The structural changes Bailey describes—deglobalization, aging, AI, supply shocks, financial system transformation—are not temporary disruptions that will revert to a familiar baseline. They are qualitative shifts that create a different economic landscape. If that is right, then the institutional adaptations described in the other three papers are not optional upgrades. They are necessary responses to a world where the past is a less reliable guide to the future.

If you only read a few pages

Read the opening framing on the five structural headwinds and the section on AI and Schumpeter’s “discrete rushes.” Then read the concluding passages on what these headwinds imply for monetary policy and international financial architecture.

The Essential: From Parameter Uncertainty to Structural Change — Brainard’s Arc

Paper: “Uncertainty and the Effectiveness of Policy” by William C. Brainard. American Economic Review, 57(2), pp. 411–425. Published 1967.

and

Chapter: “Making Policy in a Changing World” by William C. Brainard and George L. Perry. In Economic Events, Ideas, and Policies: The 1960s and After, edited by George Perry and James Tobin, pp. 43–69. Brookings Institution Press, 2000.

Most economists working on monetary policy know the Brainard principle. The 1967 paper asks a deceptively simple question: how should a policymaker act when uncertain about the effects of their own instruments? Brainard’s answer—the attenuation principle—is that when the policymaker faces multiplicative uncertainty (uncertainty about the transmission mechanism itself, not just additive noise), they should respond less aggressively than they would under certainty. Move in the right direction, but hedge against the possibility that the instrument does not work as expected.

The result is elegant, and it has shaped central bank practice for decades. The gradualism documented in the Amaral et al. survey (Paper #3) is, in substantial part, Brainard’s legacy. When central bankers say they prefer to move in measured steps because they are uncertain about the economy’s response, they are channeling—explicitly or implicitly—the 1967 paper.

But the Brainard principle rests on a crucial assumption: the policymaker knows the model. The structure of the economy is given; only the parameters are uncertain. The policy problem is to choose the right instrument setting given that you don’t know the exact coefficient. This is uncertainty within a known framework—risk, in Knight’s terminology, not genuine uncertainty about the framework itself.

Thirty-three years later, Brainard returned to the question—and the ground had shifted. In “Making Policy in a Changing World,” Brainard and Perry set out to do something that conventional econometric analysis rarely does: allow for the possibility that the key macroeconomic relationships have changed over time, and examine what policymakers knew—and didn’t know—as those changes unfolded.

Their starting point is a sharp observation about the mismatch between statistical practice and the policymaker’s problem. Standard econometrics treats parameters as constant until a structural break test rejects stability—typically requiring a t-statistic of two, or odds of twenty to one against. But as Brainard and Perry note, “twenty to one are long odds for a policymaker, for whom the costs of following the model when it is wrong can be significant.” The profession’s default—assume stability until proven otherwise—is exactly backwards from the perspective of someone who has to act under the possibility that the world has changed.

To address this, they estimate wage and price equations using Kalman filters that allow all coefficients to vary over time as random walks—not as discrete breaks at specific points, but as gradual stochastic drifts. This choice is deliberate: they argue that it is “quite implausible” that changes in wage and price processes “shift in such a discontinuous and infrequent way,” and that changes are “more likely to be spread over time.”

The results are striking. Most parameters of the inflation-unemployment relationship—the intercept, the coefficient on unemployment, the coefficient on productivity—are remarkably stable across four decades. But one crucial coefficient is not: the sum of coefficients on lagged inflation, the conventional proxy for expected inflation. This parameter—which determines whether the Phillips curve is accelerationist—was moderate in the early 1960s, rose to near 1.0 during the high-inflation OPEC shock years, and then declined back to its lowest levels by the late 1990s. The accelerationist Phillips curve, and with it the NAIRU framework, is approximately correct during high-inflation periods—but not at low or moderate inflation.

The changing coefficients in one figure

Figure 2-3 is the paper’s key exhibit. Four panels show the time-varying coefficients of a CPI equation estimated with Kalman filters from 1960 to 1998. Three of the four parameters—the intercept, inverse unemployment, and productivity—are remarkably stable across four decades. But the fourth, the sum of coefficients on lagged inflation (the conventional proxy for expected inflation), tells a different story: it rises from moderate levels in the early 1960s toward 1.0 during the high-inflation OPEC shock years, then declines steadily back to its lowest levels by the late 1990s. This single figure captures the chapter’s central finding: while most of the inflation process held steady, the one parameter that determines whether the Phillips curve is accelerationist changed dramatically—and a policymaker relying on time-invariant estimates would have been systematically misled.

Why these two papers together

The implications for policy are direct, and Brainard and Perry draw them out in their concluding section with an explicit callback to the 1967 paper. They write: “we know that uncertainty about the response to policy actions calls for deviation from certainty-equivalent behavior”—citing Brainard (1967) directly. But now the uncertainty is deeper than what the 1967 framework assumed. If parameters drift over time, then “conventional estimation procedures can be misleading” and “sticking with prior estimates, unless recent observations fall outside of conventional confidence intervals, will be a mistake.” Policymaking that is alert to parameter drift “will respond less to the prescription from conventional econometric estimates and more to recent shocks.”

The chapter closes with a sentence that could serve as an epigraph for this entire Dispatch: “policymakers need to be constantly alert to unexpected developments, both shocks and changes to the economic structure.”

The arc from 1967 to 2000 traces the same journey that central banks are making today—from parameter uncertainty to structural change. Brainard’s 1967 paper gives us the canonical framework: attenuate when uncertain. Brainard and Perry’s 2000 chapter reveals that attenuation within a known model is not enough when the model itself is shifting. The first paper asks: “What if I don’t know the coefficient?” The second asks: “What if the coefficient has changed—and I don’t know when, or to what?”

That second question is the one the four papers in this Dispatch are wrestling with. The Fed’s taxonomy distinguishes structural uncertainty from state uncertainty (Bauer et al.). The Bank of Canada invokes Knightian uncertainty by name (Cateau et al.). The BIS survey reveals that most institutions haven’t yet fully engaged with uncertainty beyond quantifiable risk (Amaral et al.). And Bailey frames the current moment as one of qualitative structural change (Bailey). Each of these is, in its own way, grappling with the problem that Brainard and Perry identified a quarter-century ago: the world changes in ways that a time-invariant model cannot capture.

Reading Brainard (1967) and Brainard and Perry (2000) together provides the intellectual foundation for this month’s theme—and a measure of how far we’ve come, and how far we still have to go.

If you only read a few pages

For the 1967 paper, read the setup and the derivation of the attenuation result—the core insight fits in a few pages. For the 2000 chapter, read the “Data and Decisionmaking” section (p. 46) for the twenty-to-one odds argument, then look at Figure 2-3 for the time-varying CPI equation parameters, and finish with “Some Lessons for the Conduct of Policy” (pp. 68–69) for the explicit link back to Brainard (1967) and the conclusions about structural change.

Closing Remarks

That’s it for the February issue of the Knightian Uncertainty Dispatch.

What emerges from reading these four pieces together is a picture of central banking at an inflection point. The taxonomy exists (Bauer et al.). At least one institution is acting on it with explicit reference to Knightian uncertainty (Cateau et al.). But the broader field remains focused on quantifiable risk (Amaral et al.)—even as the structural landscape becomes harder to map from historical data alone (Bailey).

The gap between the conceptual frontier and institutional practice is where the interesting work lies. And it is narrowing—but slowly.

If you have suggestions for papers that I should cover in future issues—especially work that connects structural change, Knightian uncertainty, and real-world forecasting and policymaking—please send them my way.

And if you found this Dispatch useful and want the next issue in your inbox, consider subscribing. It helps the Dispatch reach the people who are interested in developing economic theory and practical forecasting tools for a changing world characterized by Knightian uncertainty.

Welcome to the second issue of the Knightian Uncertainty Dispatch—a monthly curated reading list for anyone thinking about macroeconomics, finance, and forecasting in a world where the future is not just risky, but sometimes genuinely unforeseeable because it may differ from the past.

Each month, I recommend four new papers + one “essential” that help grapple with structural change and uncertainty beyond probabilistic risk—and what those realities imply for economic modeling, forecasting, and policymaking.

The goal is to curate papers that:

1. Deepen our understanding of structural change and Knightian uncertainty.

2. Are useful for how we actually reason and forecast in unstable environments.

3. Connect to one another—so the pieces speak to each other rather than living in isolated silos.

This month’s theme is forecasting in a changing world: pipelines, adaptive models, and forecast combinations.

The unifying idea is that forecasting under structural change is not about finding the true model. It is about building workflows that (i) monitor breakdown risk, (ii) adapt when relationships shift, and (iii) combine perspectives from multiple models when any single specification is likely misspecified.

The four pieces form a sequence:

• Wei documents the problem: simple constant-parameter models that used to forecast inflation well stop doing so after COVID. This is a classic example of a forecast breakdown caused by structural change.

• Hardy and Korobilis propose a new class of time-varying parameter (TVP) models where parameters adapt transparently using observable drivers. This sits between the Bianchi et al. adaptive machine learning approach and the Castle-Doornik-Hendry break-diagnostic adaptation approach discussed in the December Dispatch.

• Eraslan, Fabbri, and Saiz show how the ECB rebuilt their forecasting toolbox and evaluation pipeline post-COVID.

• Liu and Vasnev add a pluralism point: forecast combinations can induce serial correlation in errors, and you should correct for it.

Finally, this month’s essential is a “big map” of forecasting as a field and as a practice—useful background for thinking about pipelines and pluralism rather than model monism.

#1. When Inflation Forecasting Breaks: Post-COVID Evidence From the U.S.

Paper: “Forecast Breakdown of Inflation Models in the Post-COVID Era in the United States: Evidence from Classical Time-Series Methods” by Sihan Wei. Advances in Economics, Management and Political Sciences (2025).

This paper is a compact reality check on a question many forecasters lived through: models that looked respectable pre-2020 can deliver larger and more volatile forecast errors post-COVID—even for univariate models that you might have hoped were somewhat robust.

What I like about the paper is not that it discovers a surprising fact (I suspect most readers will nod along), but that it frames the episode as a forecast breakdown rather than a temporary spike in noise. In a Knightian uncertainty framing, the key issue is not only that shocks have a higher variance; it is that the data-generating process may have changed in ways that standard models did not anticipate and do not adapt to.

It is tempting to read forecast breakdown papers as a verdict on particular models (“ARIMA models failed”). But the more useful interpretation is at the workflow level: how fast do you detect that you are in a different environment, and how quickly do your forecasting models adapt? These are the questions you have to ask yourself when you take structural change seriously.

A comment: I often hear economists describe the post-2020 period as characterized by “huge shocks.” I find that phrasing problematic. In standard macro models, shocks refer to exogenous random perturbations with a fixed probability distribution (often mean-zero and independent over time). So the phrase suggests a simple rise in variance—more probabilistic uncertainty. But the kind of forecast breakdown that Wei documents points to something different: a regime change. The uncertainty is not just “more volatile” outcomes, but uncertainty about the structure itself—the hallmark of Knightian uncertainty.

(It seems to me there is an error in the labelling of the time periods in Table 2. Given the text, the numbers appear to correspond to the ordering of the rows “Pre-COVID, Post-COVID, Full,” instead of “Full, Pre-COVID, Post-COVID” as written in the table.)

Why I’m recommending it

• It documents a post-COVID forecast breakdown in a way that is easy to communicate: same model class, very different forecast error behavior after 2020.

• It emphasizes structural change rather than “bad luck,” which is the right starting point for forecasting in a changing world.

• It motivates the rest of the papers in this Dispatch: once you accept forecast breakdown, you want adaptive models (the Hardy and Korobilis paper) and model pluralism in terms of forecast combinations.

The forecast breakdown in one figure

Wei’s Figure 2 (reprinted below) shows the mean-squared forecast error (MSFE) computed over a rolling 24-month window for one-step ahead inflation forecasts, comparing standard univariate benchmarks.

The main visual message is a clear forecast breakdown: MSFEs are relatively low and stable through much of the pre-2020 sample, but they jump sharply after 2020 and remain elevated and more variable in the post-COVID period. In other words, forecast errors don’t just get temporarily noisier—they look like they move to a different level, consistent with Wei’s broader forecast breakdown interpretation and the idea that the inflation process changed in ways that standard time-series models did not anticipate.

The takeaway

Forecast instability is not an edge case. If your forecasting pipeline assumes stability, it will look great—right up until it doesn’t.

If you only read a few pages

Read the Introduction, then jump straight to the part where the paper compares pre- and post-COVID forecast errors and summarizes which models degrade most (Section 3).

#2. A Transparent Way to Make TVP-VARs Adapt Faster During Crises

Paper: “Learning from crises: A new class of time-varying parameter VARs with observable adaptation” by Nicolas Hardy and Dimitris Korobilis. arXiv preprint (2512.03763), submitted December 3, 2025.

Time-varying parameter vector autoregressive (TVP-VAR) models typically allow for structural changes by specifying their parameters to evolve as latent random walks. This paper tackles a classic practical frustration with such models: in major crises, the parameters can adapt too slowly, and in high dimensions, the model may regularize so heavily that most parameters become virtually constant.

Their proposal is the adaptively-varying parameter VAR (AVP-VAR): instead of making parameters drift purely as latent random walks, allow them to evolve deterministically as combinations of observable drivers—uncertainty indicators, stress indicators, expectations, volatility proxies, etc. That design has two attractive properties: it is more interpretable (“parameters moved because stress moved”), and it disciplines time variation without requiring a huge latent state covariance matrix. The challenge, as I see it, is how to choose and test those observable drivers.

In their forecast evaluation, the AVP-VAR on average delivers better predictive performance than standard alternatives—including conventional constant-parameter VARs and benchmark TVP-VAR specifications that rely on latent random-walk drift—and they argue (and show in simulations) that this is particularly the case when the data-generating process experiences jumps and regime shifts.

A comment: It would be quite interesting to see a plot of the absolute or squared forecast errors of the AVP-VAR compared to the constant-parameter VAR and benchmark TVP-VARs over time. That would make it easier to see when the gains materialize in the forecasting examples presented: are they concentrated in a handful of turbulent episodes, such as post-2020, or are they more diffuse across the sample?

Why I’m recommending it

• It provides a concrete mechanism for faster adaptation when relationships shift sharply—exactly the post-COVID style environment highlighted by Wei.

• It links the time-variation in the parameters directly to movements in observable series, instead of letting it be driven by an opaque latent innovation process.

• It addresses a practical feature of canonical TVP-VARs: over-regularization that leaves time-varying parameters nearly flat during the very episodes where you want movement.

The forecast evaluation in one table

Table 2 from the paper is reproduced below. It reports out-of-sample forecast performance across horizons and models, benchmarked against a constant-parameter VAR. The forecasted variables are monthly U.S. data for industrial production (Panel A) and the personal consumption expenditures price index (PCEPI; Panel B), spanning January 1985 to November 2023.

I have highlighted the AVP-VAR column to make the comparison immediate. The pattern is clear: the AVP-VAR delivers lower mean squared forecast errors on average than the constant-parameter VAR and TVP-VAR benchmarks across most horizons (and in both panels), consistent with the paper’s central claim that observable adaptation improves forecast performance.

The takeaway

In their out-of-sample horse race, the AVP-VAR outperforms standard VAR and TVP-VAR benchmarks across most horizons—evidence that “observable adaptation” can translate into real forecast gains, not just nicer stories.

If you only read a few pages

Read the Introduction for the motivation (TVP-VARs adapt too slowly and regularize too hard), then go to the sections that define the AVP-VAR structure (Section 2.2) and the pseudo out-of-sample forecast comparison against standard VAR/TVP benchmarks (Sections 3.2 and 3.3).

#3. How the ECB Rebuilt Its Short-Term GDP Forecasting Toolbox Post-COVID

Paper: “Short-term forecasting of euro area economic activity in an uncertain world” by Sercan Eraslan, Andrea Fabbri, and Lorena Saiz. ECB Economic Bulletin, Issue 8/2025.

This article explains how the ECB updated and stress-tested its short-term GDP forecasting framework after COVID—a period that “significantly disrupted the performance of traditional forecasting methods.”

Concretely, they rebuilt the toolbox around two ideas: (i) refresh and broaden the set of nowcasting models used for short-term GDP, and (ii) make the pipeline more resilient to the post-2020 data environment. The updated framework combines bridge equations with alternative signal-extraction approaches—dynamic factor models and VAR/MIDAS-style specifications—and it adds a machine-learning cross-check (a Quantile Regression Forest) to cope better with non-linearities and the unusual mix of volatility and outliers that characterized the post-COVID period.

On performance, the headline is pragmatic rather than triumphant: the revised toolbox improves short-term forecasting on average and, just as importantly, reduces the risk of being misled by any single model when relationships are unstable. The ML cross-check is particularly useful as a diagnostic—less as a replacement for the workhorse models than as a way to flag when the standard models may be extrapolating from a world that no longer looks like the present.

A comment. One passage at the end of Section 2 is especially revealing: the authors note that structural change is outside the scope of the revisions they have implemented. Instead, the revisions focus on time-varying volatility and outlier treatment, while the core models still rely on constant parameters estimated on a long sample starting in the mid-1990s. That is understandable in an institutional setting, but it points to something deeper: much of macro theory, econometrics, and forecasting still takes stable structure as the default, with “instability fixes” layered on top. What we arguably still lack is a widely adopted, unifying framework that treats structural change as foundational—and that links economic reasoning, empirical modeling, and operational forecasting around a shared, instability-aware view of the economy.

Why I’m recommending it

• It is a rare window into how a major policy institution operationalizes forecasting under instability: toolbox design, evaluation discipline, and cross-checks.

• It shows how different models are used for different tasks: some translate high-frequency indicators into a nowcast, others to extract common signals and handle data-rich settings, others to capture dynamic interactions and mixed frequencies. And the machine learning model acts as a complementary non-linear cross-check.

• It contains a useful tension: even as the ECB improves robustness to volatility and outliers, it explicitly brackets structural change—highlighting how persistent the constant-structure mindset is, and why we still need forecasting frameworks built around structural change.

Improved forecast performance in one graph

Chart 1 (reprinted below) provides a simple before/after comparison of the ECB’s short-term GDP forecasting performance. It plots two standard diagnostics—Bias (systematic over- or underprediction) and MAFE (mean absolute forecast error)—for the old versus the revised framework.

The message is reassuringly practical: the updated toolbox delivers forecasts that are less systematically off target (bias closer to zero) and more accurate on average (lower MAFE). The improvements are not framed as a dramatic “model revolution,” but as a cumulative payoff to strengthening the pipeline. It would, however, be interesting to see how the forecast performance has evolved over time.

The takeaway

Forecasting under instability is less about upgrading a single model and more about upgrading the workflow: maintain multiple methods, refresh them systematically, and use cross-checks to detect when the baseline is extrapolating from an outdated structure. At the same time, the paper is a reminder that many “robustness” upgrades still live inside a constant-parameter worldview.

If you only read a few pages

Read Section 2 for the motivation and then Section 3 for the revised toolbox. Box 2 is particularly useful because it explains the details of the new forecasting model.

#4. A Subtle Pitfall in Forecast Combinations—and a Fix

Paper: “Corrected Forecast Combinations” by Chu-An Liu and Andrey L. Vasnev. arXiv preprint (2601.09999), January 16, 2026.

Forecast combinations are a natural response to model uncertainty and structural change: when you don’t trust a single specification, you average across them. But this paper points out a practical wrinkle: even if individual model errors look well-behaved, the combined forecast errors can become serially correlated—and that autocorrelation matters for how you estimate optimal weights and evaluate forecast performance.

Their contribution is to propose to “correct” the forecast combination with a simple error-correction term—adding a weighted lag of the previous combined forecast error—so the combination explicitly absorbs serial dependence rather than assuming the combined errors are white noise.

Why I’m recommending it

• It shows that forecast combinations are not a free lunch: the combined forecast errors can be serially correlated, and ignoring that can distort both weighting and evaluation.

• It offers a correction that is both statistically motivated and operationally simple: one extra term (a lagged error) can remove a lot of serial dependence without turning the combination into a complicated black box.

• It highlights an important point for pipeline design: if you combine forecasts, you should add ensemble-level diagnostics (and, if needed, simple corrections) rather than treating the combination as a black box.

The forecast combination correction in one figure

Figure 4 (reprinted below) makes the correction idea very concrete by plotting the autocorrelation of forecast errors for different combination methods. The basic diagnostic is in panel (a): the forecast errors from the simple mean combination are clearly autocorrelated at multiple lags—so part of what looks like “forecast errors” is actually predictable structure.

Then the fix is visible immediately in panel (b): once they apply the correction—i.e., correct today’s combined forecast by adding a weighted lag of last period’s combined error (here using a correction factor of 0.65)—the remaining forecast errors look much closer to white noise, with the autocorrelation largely stripped out.

The same logic shows up even more starkly for “optimal” combinations. This is clear from panels (c) and (d). Finally, panel (e) shows that if you estimate the weights and the correction factor jointly by GLS, you get a similar result: the correction soaks up the predictable component of the combined errors.

In short, Figure 4 nicely visualizes that “correction” is not just cosmetic—it directly targets a measurable failure of forecast combinations: serially dependent errors.

The takeaway

Forecast combination is powerful under structural change—but treat the combination stage as part of the model, with its own diagnostics and corrections.

If you only read a few pages

Read the Introduction and the core setup (Section 2) that shows how serial dependence can emerge in combined errors. Then jump to the empirical illustration.

The Essential: A Big Map of Forecasting as Practice

Paper: “Forecasting: theory and practice” by Fotios Petropoulos and many others. International Journal of Forecasting. 2022.

This is not a paper you read cover-to-cover in one sitting. It’s a field guide: a broad review of forecasting models, methods, and practice—how people actually build, organize, evaluate, and deploy forecasts.

I’m including it as the Essential for this issue because it matches the theme perfectly: forecasting in a changing world is about workflows and pluralism. This paper makes that idea concrete by treating forecasting as an end-to-end activity, not as a single model selection problem.

Why I’m recommending it

• It is explicitly pipeline-shaped: forecasting is presented as preparation, production, organization, and evaluation—not just estimation.

• It normalizes pluralism: the diversity of applications implies a variety of methods; no single approach dominates across forecast contexts.

• It is rich on evaluation and practice, which is where structural change shows up most painfully (breakdowns, revisions, instability).

The takeaway

If you want to forecast in an unstable world, you need to think like a system designer: methods + evaluation + monitoring + communication. This paper is one of the best “maps” for that mindset.

If you only read a few pages

With a total of 167 pages, recommending only a few to read is quite difficult. Instead, treat the paper as a reference you return to when designing or updating your forecasting pipeline.

That’s it for the January issue of the Knightian Uncertainty Dispatch.

If you have suggestions for papers that I should cover in future issues—especially work that connects structural change, Knightian uncertainty, and real-world forecasting workflows—please send them my way.

And if you found this Dispatch useful and want the next issue in your inbox, consider subscribing. It helps the Dispatch reach the people who are interested in developing economic theory and practical forecasting tools for a changing world characterized by Knightian uncertainty.

Welcome to the Knightian Uncertainty Dispatch—a monthly curated reading list for anyone thinking about macroeconomics, finance, and forecasting in a world where the future is not just risky, but sometimes genuinely unforeseeable because it may differ from the past.

Each month, I’ll recommend new research papers and one essential that help grapple with structural change and uncertainty beyond probabilistic risk—and with what those realities imply for economic modelling, forecasting, and policymaking.

The goal is simple: curate papers that (i) deepen our understanding of uncertainty and structural change and their implications, (ii) are useful for how we actually reason and forecast in a changing economic environment, and (iii) connect to one another, so the pieces speak to each other rather than living in separate silos. I’ll keep the blurbs short, explain why I think each paper is worth your time, point you to key figures and sections, summarize the main takeaways, draw links across papers, and add a comment or two where interpretation matters.

I hope you find this useful.

#1. How the Bank of England Deals With Uncertainty

Paper: “Monetary policymaking under uncertainty” by Alex Haberis, Richard Harrison, Kate Reinold, and Matt Waldron. Published on October 17, 2025.

This new Macro Technical Series paper from Bank of England staff is one of the clearest explanations I’ve seen of how a modern central bank uses models to support monetary policymaking in a world of pervasive and time-varying uncertainty—much of which is not captured by standard economic models.

(I wrote a full post about this paper a few weeks ago.)

I recommend it if you’re interested in how policy is actually conducted when economic models are misspecified, the economy changes, and the playbook has to adapt—especially if you think of uncertainty as Knightian uncertainty arising from unforeseeable change, rather than just probabilistic risk.

Why I’m recommending it:

• They draw a practical distinction between resolvable uncertainty and epistemic/unresolvable uncertainty, stressing that policymakers typically can’t write “the true” joint distribution of outcomes. That’s exactly the conceptual territory that motivates my work on Knightian uncertainty.

• They explicitly connect “pervasive and time-varying” real-world uncertainty to structural change, noting that existing theoretical approaches are often too simplified to be directly applicable.

• They organize sources of uncertainty in a way that’s immediately useful for macro, finance, and forecasting conversations: uncertainty about data/measurement, shocks/events, and transmission/structure.

The time-varying uncertainty in one figure

The figure below gives a nice visual illustration of why uncertainty is not one thing: recent episodes span everything from probabilistic risk (upper-left corner, A) to deep uncertainty (lower-right corner, C).

The takeaway: three complementary frameworks

The paper emphasizes that policymaking in the “grey area” of uncertainty is about disciplining judgment, not replacing it. They do so by combining three complementary perspectives:

1. Forecast-based: Use quantitative projections, but loosen the tight “model → forecast → policy” linkage by combining multiple models with judgment when the environment is shifting.

2. News-based: Treat policy as a sequence of updates driven by an evolving interpretation of shocks and mechanisms—particularly useful when forecasting is intrinsically hard (for example, in periods with major structural changes).

3. Rules-based: Use simple, transparent, and reasonably robust rules as benchmarks—not to mechanize decisions, but to anchor deliberation and communicate systematically.

If you only read a few pages

Skim the early framework discussion—especially the distinction between resolvable vs. epistemic uncertainty (Section 2) and the summary of the “three perspectives” (page 18)—then come back to Figure 1, which places recent episodes along the spectrum from probabilistic risk to deep uncertainty, and Figure 3, which shows the usefulness of the three perspectives across this spectrum.

#2. Quantifying a Time-Varying Bias in Survey Expectations of Stock Returns and Earnings

Paper: “The Prestakes of Stock Market Investing” by Francesco Bianchi, Do Lee, Sydney C. Ludvigson, and Sai Ma. Working paper. Latest version: December 2, 2025.

This new working paper quantifies the “bias” in survey-based expectations by comparing them to a clever benchmark: a real-time, data-rich, out-of-sample machine-learning forecast that is explicitly designed to adapt to a changing economic landscape.

I’m recommending it because it addresses a big limitation of a lot of empirical work on survey expectations: it often does not account for structural change in the relationships that link information to outcomes and expectations.

A comment

I disagree with the authors’ label of their benchmark forecasts as “objective beliefs”. The benchmark is not objective; it is an impressive adaptive forecasting algorithm built to satisfy particular criteria chosen by the authors. More importantly, once we take unforeseeable structural change seriously, a systematic deviation of survey expectations from any benchmark does not imply irrationality. (See my post on Rational Expectations Under Knightian Uncertainty.)

Why I’m recommending it:

• They construct a benchmark forecast of stock returns and corporate earnings using a machine-learning algorithm that produces true out-of-sample forecasts using only information that would have been available at the time (and in a very data-rich way).

• The algorithm is designed to adapt, including by tuning training/validation window lengths, because no single fixed historical window is optimal when the economy may be subject to slow-moving shifts or breaks. This kind of design is crucial for forecasting under structural change.

• They compare the resulting time-varying benchmark forecasts to survey expectations and show that, in key episodes, survey expectations vary too little and do not adjust quickly enough to a changing economic landscape.

The estimated time-varying bias in one figure

The figure below shows the ex post survey forecast errors for corporate earnings and stock returns (actual outcomes minus the survey forecasts; black lines) and decomposes these errors into a “bias” (survey forecast minus benchmark forecasts; blue bars) and an “ex post error” (benchmark forecast minus actual outcome; red bars). The large blue bars around 2008 and after 2020 are particularly striking.

The takeaway: a structural-change-aware benchmark forecast + survey expectations adapt too little

The key methodological contribution is the paired design: a machine-learning forecast uses real-time information to produce optimal forecasts that can adapt to structural change. The survey-benchmark gap then provides a time-varying measure of the bias (the “prestakes) in the survey expectations. That said—as I argued above—this bias should not necessarily be interpreted as evidence that survey respondents are irrational.

Empirically, the main result is that the machine-learning forecasts move and “course-correct” sharply around major turning points (such as the Great Financial Crisis and the post-COVID period), while survey expectations adjust too little—particularly in turbulent periods—so the largest measured biases show up exactly when the stakes are highest.

This emphasizes the importance of forecasting methods adapting to structural change, and it illustrates that cleverly designed algorithms can do this better than survey respondents have done.

If you only read a few pages

Read the Introduction and Section 2’s construction of the machine-learning benchmark (particularly the discussion of how it is designed to handle changing environments, pp. 9-10), then jump to the empirical results in Figure 2.

#3. Rapidly Adjusting Forecasts After Structural Breaks

Paper: “A Novel Approach to Forecasting After Large Forecast Errors” by Jennifer L. Castle, Jurgen A. Doornik, and David F. Hendry. Published with open access in the Journal of Forecasting, October 2025.

This paper proposes an idea that runs through Prof. David Hendry’s approach to econometrics and forecasting (which I am hugely influenced by): when the world changes, your model will start producing a run of large, same-sign 1-step-ahead forecast errors—those errors contain actionable information about what kind of break you are facing. The authors turn that idea into a simple, fast, and interpretable procedure for getting forecasts “back on track” after structural breaks.

I’m recommending it because it’s one of the most concrete “do this tomorrow” forecasting fixes I know for dealing with structural change in time-series models—and it’s fully in the spirit of this Dispatch: treat instability as the rule, not an edge case.

Why I’m recommending it:

• The approach is explicitly designed to handle structural breaks: a short sequence of one-sided forecast errors is interpreted as evidence of an unanticipated shift, not noise.

• It’s interpretable: rather than letting a black box absorb the instability, the method tests whether the forecast failure is best explained by an outlier/measurement error, a level (step) shift, or a broken trend. The forecasting model is then updated accordingly.

• It uses indicator saturation + model selection (using Autometrics in OxMetrics) to detect shifts quickly at the end of the sample period, with tight significance levels to avoid spurious selection of breaks.

The approach in one figure

Figure 7 (reprinted below) captures the entire approach at a glance. In panels (a) and (b), the red line shows the realized inflation series, while the blue dotted lines show one-step ahead out-of-sample forecasts based on estimation samples ending in 2021(3) and 2021(4), respectively. The growing gap between the red series and the blue forecasts in the final observations corresponds to large, positive forecast errors.

Panel (c) shows what happens when the sample period is extended to 2021(5): the algorithm selects a broken (log) trend and adds it to the model. This shifts the out-of-sample forecasts upwards—from the range marked by the vertical black line to the higher range marked by the vertical purple line. Finally, panel (d) shows how payoff: by including the broken trend, the method adapts quickly to the structural break, producing post-break forecasts that are materially closer to the realized inflation path than the forecasts that omit the break.

The takeaway: “large forecast errors are signals” + a disciplined adaptation

Their approach automatically and systematically captures the procedure:

1. When you observe a big 1-step forecast error, test whether to include an impulse indicator as an intercept correction to reset the forecast origin;

2. If a second (or third) large, same-sign forecast error follows, test whether the intercept correction can be replaced by a step or broken linear/log-linear trend;

3. Remarkably, even with very few post-break observations, the selected broken trend can forecast acceptably well—until the next break.

They illustrate this on UK inflation during 2021-2024: repeated forecasting failures around the post-COVID reopening and the energy shocks are used to detect successive breaks and update the forecasting model, with episodes where the updated device tracks the surge and subsequent decline much better than “do nothing” forecasting.

Links to other papers in this Dispatch

This paper also pairs beautifully with the Bianchi et al. paper I recommended above: both are about adapting forecasts under structural change, but they rely on different approaches. Bianchi et al. build an adaptive forecast using machine learning. Castle, Doornik, and Hendry show how to adapt using transparent break diagnostics (impulses/steps/broken trends) that you can interpret and communicate.

If you only read a few pages

Read the part that explains the three-step correction loop (intercept correction → test step/broken trend → keep forecasting until next break) in Section 3, then skim the UK inflation application figures in Section 6 to see how quickly the method detects and adjusts to successive shifts.

#4. Could the Bank of England Have Avoided Mis-Forecasting Inflation in 2021-24?

Paper: “Could the Bank of England have avoided mis-forecasting UK inflation during 2021–24?” by Jennifer L. Castle, Jurgen A. Doornik, and David F. Hendry. Forthcoming in International Journal of Forecasting, 2026. Open access.

This paper considers the post-2021 UK inflation episode and asks a sharp, practical question: once inflation started to surge and forecast errors began to stack up, could the Bank of England have updated its forecasts faster and more effectively? Their answer is essentially “yes”—not by guessing shocks in advance, but by treating successive large, same-sign one-step forecast errors as information about a trend break, and using that information to update forecasts rapidly.

I’m recommending it because it’s a clean bridge between (i) the reality of policymaking under uncertainty partly arising from structural change (the Bank of England staff paper I recommended above) and (ii) a concrete, implementable approach to adapting forecasts under structural breaks (the previous Castle, Doornik, and Hendry paper I recommended above).

Why I’m recommending it:

• It’s explicit about what is—and isn’t—forecastable: unpredictable shocks explain some failures, but the paper argues that tardy reactions to accumulating evidence also mattered.

• It operationalizes structural change in real time: two (sometimes three) large, increasing, same-sign one-step forecast errors are used to estimate broken linear or log-linear trends quickly, even with very few post-break observations.

• It puts numbers on the comparison: their adaptive approach yields lower root mean squared forecast errors (RMSFEs) over 2021(4)-2024(3) than the BoE’s Monetary Policy Report projections at comparable horizons (Table 2).

The forecast comparison in one figure

Figure 3 (reprinted below) is the “policy-relevant” illustration: it overlays the BoE’s four-times-a-year Monetary Policy Report projections with the authors’ adaptive forecasts and the realized inflation path, making it easy to see when the BoE’s projections lagged the turning points and how the adaptive updates reduce the systematic miss.

The takeaway: monitor continuously + update when the trend shifts

The paper’s forecasting philosophy is simple and directly “structural break aware”: don’t forecast with fixed models; instead, monitor one-step errors as the data arrive, and when they start failing systematically, update the forecasting model immediately.

They implement this with a very simple univariate trend model for the log price level, using indicator saturation to handle past shifts, then adding intercept corrections and—when the forecast errors persist—broken trends to capture the new regime.

Link to other papers in this Dispatch

• To the previous Castle, Doornik, and Hendry paper: same approach (intercept correction → test step/broken trend → keep forecasting until next break), but here the focus is on the direct comparison with the BoE’s published projections.

• To the BoE staff uncertainty paper: it’s a nice complement. The BoE staff paper explains how policy is conducted under pervasive and time-varying uncertainty; this paper asks what a structural-breaks-aware forecasting workflow could add to the Bank’s toolbox in exactly the kind of episode where models “fall apart.”

If you only read a few pages

Read the Introduction and Section 2 for the core idea (large same-sign errors contain information about breaks), then go to Section 6 and focus on Figures 1-3 and Table 2 for the real-time update logic and the head-to-head comparison with the BoE’s projections.

The Essential: A Survey of Forecasting in the Presence of Instability

Paper: “Forecasting in the Presence of Instabilities: How We Know Whether Models Predict Well and How to Improve Them” by Barbara Rossi. Published in the Journal of Economic Literature, 2021.

This is my “essential” this month because it’s a clear, practical guide on how to evaluate and improve forecasts in the presence of structural instability.

Rossi’s central message fits the theme of this Dispatch perfectly: what matters is not only whether your model looks stable in-sample, but whether it delivers stable, satisfactory forecasting performance out of sample in the presence of structural change.

Why I’m recommending it:

• A core (and surprisingly underappreciated) point: breaks in a model’s parameters are neither necessary nor sufficient for time-variation in forecasting performance—so you shouldn’t “solve” instability by only testing for structural breaks in-sample. You should also evaluate forecasting ability in ways that are robust to instabilities.

• Rossi argues for local diagnostics of forecasting performance—because full-sample average diagnostic tests can miss the exact pattern we care about (e.g., offsetting under- and over-prediction across regimes).

• Rossi argues that reporting forecast uncertainty (fan charts, densities, calibration) is crucial—especially for central banks—and shows why standard reporting of point forecasts can be misleading when the environment is unstable.

The takeaway: evaluate forecasts locally + adapt deliberately + report uncertainty honestly

Rossi organizes the forecasting problem around four questions: what instabilities are, how to evaluate forecasts under them, how to improve forecasts under them, and how to measure and report forecast uncertainty under them.

For me, the “toolkit” takeaway is:

• Evaluation: Use forecast evaluation tests designed to detect time variation in performance (not just average performance). Table 1 is a great map from “traditional tests” to “robust-to-instability tests.”

• Improvement: Don’t wait for a perfect break model. Use practical, real-time strategies that trade off bias and variance—intercept corrections, rolling/discounted estimation, window selection/combination—plus “big data” approaches and forecast combinations as insurance against “pockets of predictability.” Table 2 is the quick overview.

• Reporting: In unstable environments, you need predictive densities and careful calibration checks; fan charts are not automatically informative unless their construction respects instabilities and asymmetries. Rossi also notes that “best critical regions” can induce robustness to extreme events that cannot be quantified, linking naturally to Knightian uncertainty.

Links to the other papers in this Dispatch

• It’s a conceptual “umbrella” over the Castle-Doornik-Hendry adaptive forecasting papers: those papers are essentially specific implementations of the broader guidance Rossi surveys (monitoring breakdowns, adapting quickly, and evaluating performance locally rather than on average).

• It also complements the Bank of England’s staff paper: Rossi’s message is that instability is ubiquitous and forecast performance is time-varying—and that institutions should use evaluation and communication tools that don’t get fooled by regime averaging.

If you only read a few pages

Read the Introduction and the four-question roadmap, then go straight to Table 1 (robust evaluation tests) and Table 2 (forecast-improvement strategies). After that, skim the discussion about intercept corrections and window choice in Section 4.1.1 for actionable adaptation ideas, and the beginning of Section 5 about the predictive densities and fan charts for how to think about forecast uncertainty reporting under instability.

That’s it for this month’s Knightian Uncertainty Dispatch. If you read (or write) papers that belong in future issues—new work, overlooked gems, or “essentials” that deserve a slot—please don’t hesitate to comment below or reach out with your suggestions. I’m always looking for pieces that take uncertainty, model misspecification, and genuinely unforeseeable change seriously, and I’d love to hear what you are reading.