How the Federal Reserve Made ‘What If?’ Institutional
The Federal Reserve's scenario practice is best understood as an institutional lifecycle, not as a growing pile of forecast tables. This chapter traces what persisted as the practice moved from a crisis-era exercise to a recurring part of FOMC materials, and how changes in models expanded the risks the staff could represent. The useful judgment is to separate greater analytical capacity from greater certainty: a richer scenario can expose more realistic feedbacks without becoming a more reliable prediction.
From crisis exercise to recurring FOMC practice
The first known public mention of “alternative scenarios” in the Federal Reserve's FOMC historical materials came at the April 1968 meeting. The setting was unusually fragile: the Gold Pool had collapsed, and policymakers were dealing with arrangements intended to manage the private-sector price of gold. Robert Solomon presented two paths for the period ahead. One assumed a return to international monetary stability through restored confidence that the existing gold price would hold. The other assumed renewed instability if the United States failed to correct its balance-of-payments problem.
This was already the essential form of scenario analysis. The staff did not need to pretend that both paths were equally likely. It needed to show that the policy consequences depended on a condition that might resolve in one of two ways. The scenario made the condition explicit, linked it to an international mechanism, and gave the FOMC a basis for considering how monetary policy would differ across the branches.
The historical study is careful about what this beginning means. It calls 1968 the first known mention in publicly available FOMC materials, not necessarily the first scenario ever prepared inside the Federal Reserve. That distinction matters because institutional history is always partly a history of surviving records. A public document can show what was presented or documented; it cannot establish the absence of other internal analysis.
The practice changed state as the Federal Reserve accumulated both a larger documentary record and more modeling capacity. In the later decades, scenarios appeared in the written materials prepared for FOMC meetings and in material discussed during the meetings. The study identifies 1,265 staff alternative scenarios from 1968 through 2020 across publicly released Greenbooks, Bluebooks, Tealbooks, transcripts, memoranda of discussion, and presentation materials. That number describes the researchers' assembled public corpus, not a claim that every internal thought experiment was preserved or that every item had the same level of detail.
A structured subset makes the change in practice more visible. From 1989 through 2020, 1,132 scenarios appeared in Greenbook and Tealbook A materials with narrative descriptions and numerical forecasts for selected macroeconomic variables. In a typical Greenbook or Tealbook A in the first two decades of the twenty-first century, the staff presented about six or seven alternatives. Meeting materials could refer to those paths or add others, reaching as many as ten scenarios in an FOMC round.
The important transition is not simply from fewer scenarios to more. An occasional exercise can call attention to a crisis. A recurring practice must also support comparison: the same broad baseline, a recognizable way to describe the departure, and enough consistency for policymakers to understand what has changed. The Federal Reserve's later materials made alternatives more legible as a set around a central projection. They remained shorter and less resource-intensive than the baseline, but their repeated presence made risk discussion part of the ordinary architecture of deliberation rather than an exceptional response to a breakdown.
Regularity changes the institutional question. Once alternatives recur, staff can compare not only each branch with the baseline but also the kinds of risks that are repeatedly being considered. A scenario set can cover different directions for output and inflation, while the baseline continues to say which path is currently judged most likely. That arrangement gives the FOMC a common map for discussing disagreement: participants can challenge the assumed shock, the modeled transmission, or the judgment about which branch deserves attention, rather than arguing only over a single central number.
The transition also introduces a discipline problem. More routine production does not mean that every possible risk is represented. A scenario is still a selection, and a recurring format can make the selected categories look more complete than they are. The public record shows expansion and standardization; it does not show that the Federal Reserve had a full inventory of future threats. Institutionalization makes the “what if?” question repeatable, but it does not make the unknown known.
A second transition was the connection between numerical paths and qualitative judgment. Recurring staff assessments of forecast uncertainty and the balance of risks began appearing alongside the scenario material in the later sample. The inaugural Tealbook in June 2010 consolidated alternative simulations in a dedicated Risks & Uncertainty chapter, while risk assessments had begun appearing in the Greenbook in 2008 and were included in every Tealbook from March 2013 through the end of the sample.
That pairing preserved two different kinds of information. A scenario could show what would happen under a specified change. A qualitative assessment could describe the staff's broader view of uncertainty and risk without requiring every concern to be turned into a numerical path. The first is a conditional mechanism; the second is a judgment about the surrounding field. The Federal Reserve did not make one mechanically replace the other. The institutional gain was the ability to keep both visible in the same policy conversation.
This separation is especially useful when the institution's confidence concerns the range of outcomes rather than one particular branch. A numerical path answers a “what if this happens?” question. The qualitative assessment can say that the surrounding outlook is unusually uncertain, or that risks do not point evenly in both directions, without pretending that a small number of scenarios measures the full distribution. The two forms of evidence can disagree without one being an error: the staff might see a broad field of uncertainty while choosing only a few branches for detailed modeling.
The practice also retained a stable reference point through its transitions. The baseline remained the modal or most likely projection, supported by a more comprehensive analysis. Alternatives continued to derive their meaning from their departure from that baseline. What changed was the range of questions the staff could ask, the number of variables through which it could express the answer, and the regularity with which the questions were brought to policymakers.
Models changed what the Federal Reserve could ask
Model development changed the scenario practice because it changed the mechanisms that could be represented. The first alternative scenarios coincided with the Federal Reserve's first large-scale econometric model, the MIT-Penn-SSRC model. In the late 1970s, the Federal Reserve's Multi-country Model made a wider set of international and exchange-rate scenarios possible. The two models were often used together as the “Board model” in the 1980s and early 1990s.
The point of this history is not that one model suddenly made the economy transparent. Early models had limitations, and staff and FOMC participants sometimes recognized that the numerical results were imperfect. A model nevertheless supplied a common translation device. It allowed an assumption about fiscal policy, an exchange rate, or another disturbance to be converted into a path for growth, unemployment, and prices, rather than left as an unstructured warning.
That translation device also made assumptions contestable. A policymaker could ask whether an exchange-rate move would persist, whether households would change spending, or whether firms would pass higher costs into prices. The model's output did not settle those questions, but it exposed where they entered the path. Scenario analysis became more useful when the assumption and the response could be separated: disagreement about the starting shock was not the same as disagreement about the economy's sensitivity to it.
The July 1996 launch of FRB/US, with an improved treatment of expectations, enabled a rapid expansion in the number of scenarios presented to the FOMC. Expectations matter because households, firms, investors, and policymakers respond not only to current conditions but also to what they think will happen. A scenario that can represent those anticipations can ask a more consequential question than “what if a variable moves?” It can ask how beliefs about future income, inflation, or policy amplify or dampen the initial change.
This matters for the timing of a risk as well as its size. If people expect an inflation shock to persist, they may change wages, prices, spending, or asset allocations before the full shock appears in the data. If they expect the central bank to offset it, the same initial disturbance may transmit differently. A model that represents expectations can make those feedbacks explicit, while still leaving the underlying behavioral assumptions open to challenge.
In the 2010s, the staff increasingly used New Keynesian dynamic stochastic general equilibrium models, including SIGMA and EDO, alongside other models and event-driven extensions. These models gave scenarios a way to represent interactions among demand, supply, inflation, expectations, and policy over time. The practice also used multiple models when one framework could not carry the relevant mechanism. During the pandemic, for example, a December 2020 alternative simulation combined a small New Keynesian model with an SIR model of viral propagation so that the endogenous response of consumption and production to the spread of the virus could be represented.
This is a change in what “a shock” means. In a simple exercise, a parameter may be shifted and the consequences calculated. In a richer exercise, the same disturbance can alter behavior, financial conditions, or the information available to policymakers, which then feeds back into later decisions. The scenario becomes more realistic because it contains more links. It also becomes more dependent on assumptions about those links. Added structure is an opportunity to expose a mechanism and a new surface on which the mechanism can be wrong.
The extra links are valuable only when they answer the question the scenario was meant to ask. An international model can clarify how exchange rates or foreign demand transmit a disturbance, but it does not automatically improve a domestic productivity scenario. A learning mechanism can clarify policy error when the state of the economy is uncertain, but it adds little if the exercise assumes perfect information from the start. The choice of model is therefore part of the scenario's meaning, not a neutral technical wrapper around the result.
The Federal Reserve's examples make that progression concrete. An August 1995 natural-rate scenario mechanically moved the natural rate of unemployment up or down by 0.5 percentage points and assumed policymakers knew the true value. By January 2016, a natural-rate scenario included noisy signals, learning, and gradual discovery of the true rate, producing persistent policy errors while policymakers tried to infer the state of the economy. The later exercise did not merely use a more elaborate number. It represented uncertainty about knowledge itself.
Financial scenarios developed in a similar direction. A November 1999 scenario involving a 20% stock-price decline primarily presented a simple wealth effect. A June 2018 financial-correction scenario included higher risk premiums, curtailed credit, sentiment shocks, and a policy response, including a return to the effective lower bound. The mechanisms matter because a fall in asset prices can affect spending directly, but it can also change borrowing conditions, confidence, and the ability of monetary policy to respond.
The practice eventually modeled uncertainty about economic structure as well as uncertainty about events. A June 2002 scenario simply posited a different value for the long-run equilibrium real federal funds rate, often written as r-star. Scenarios beginning in March 2015 modeled gradual recognition that this rate had changed. In the first formulation, the unknown is the value. In the second, the unknown also changes what policymakers believe and when they learn it. That distinction can generate a persistent gap between the policy that would be appropriate with perfect knowledge and the policy chosen under imperfect information.
The reporting surface expanded with the analytical surface. Earlier briefing materials generally contained fewer scenarios, shorter horizons, and fewer variables. Over most of the 1989–2020 period, the structured tables commonly reported real GDP growth, the unemployment rate, and a measure of core inflation. The federal funds rate was added in 2006, and total PCE inflation followed in 2011. Longer horizons and more variables made it easier to examine how a risk unfolded and how policy might respond, but they also extended the chain of assumptions. More columns do not make the future more certain; they make more of the conditional story inspectable.
A longer horizon can reveal persistence that is invisible in the first few quarters. Learning about a changed natural rate, for instance, can produce policy errors that continue after the initial signal because policymakers update their beliefs gradually. A financial disturbance can also move from asset prices to credit availability and then to spending. Showing those stages helps the FOMC ask whether a risk is temporary or self-reinforcing. It does not establish that the model has captured every stage that would matter in reality.
The mature practice can therefore be pictured as a sequence of institutional states:
What persists across every state is the relationship to the baseline: a changed assumption is translated into a conditional path so policymakers can examine a risk and its consequences. What changes is the mechanism the path can carry. International linkages, expectations, financial amplification, learning, structural change, and pandemic dynamics become available as questions rather than remaining outside the model's frame.
That growth should not be mistaken for a guarantee of accuracy. The Federal Reserve paper describes a public historical record through 2020, and its classifications were ultimately produced through human judgment after AI-assisted coding. The record can establish how scenarios were presented, what risks they emphasized, and how their paths compared retrospectively with outcomes. It cannot by itself show every internal deliberation, prove that a more sophisticated scenario was better in advance, or establish what the Federal Reserve's practice became after the sample ended.
The institutional lesson is more restrained and more useful. The Federal Reserve made “what if?” a repeatable policy question by preserving a baseline, adding structured alternatives, and expanding the mechanisms through which risks could be examined. The next step is to read those alternatives by risk family—to see why a demand shock, a supply shock, a financial disturbance, or a change in expectations can produce different output–inflation trade-offs even when they begin with a similar concern.