The “Oven” of Macroeconomic Expectations: What Do We Know So Far?

Households form expectations about the evolution of economic outcomes when making decisions regarding consumption, saving, and indebtedness. This process is not exclusive to economists, although the methods used may differ. Analyzing how households form expectations is not merely an academic exercise; rather, it is essential for better understanding the dynamics of business cycles and the transmission of macroeconomic policy. In the United States, private consumption accounted for 68.5% of nominal GDP in December 2025. Consequently, more than two-thirds of output depends largely on household decisions (CEIC Data, n.d.).

Whereas professional forecasters tend to use statistical and econometric tools to project variables such as output, inflation, and unemployment, households make decisions based on the signals they receive from their surroundings. For example, the recent oil shock associated with the Iran conflict has affected countries around the world, and information about its economic impact has been conveyed not only through more formal and aggregated sources, such as central banks and governments, but also through local and informal channels, such as social media and conversations with peers.

In my view, expectation formation can be thought of as an oven: signals serve as the ingredients (information inputs), while individuals’ mental models of the economy, including their perceptions of how variables interact, correspond to the oven settings (such as temperature, timing, and conditions). The resulting expectations about future outcomes are the final dish.

Academic research introduced the concept of rational expectations (Muth, 1961), which posits that individuals use all available information and understand the true structure of the economy when forecasting variables. When shocks occur, agents update their beliefs accordingly and, on average, do not make systematic errors. In the cooking analogy, this is akin to knowing all available ingredients, both at home and in supermarkets (including all brands), as well as all existing recipes for cakes. Consequently, after cooking, one should obtain a cake that closely resembles the outcome described in the recipe.

Over time, alternative approaches to expectation formation have emerged. Assuming that individuals know and process all available information is unrealistic; even if such information were accessible, processing it is cognitively and time intensive, particularly in an environment characterized by vast amounts of data. For these reasons, the concept of rational inattention (Sims, 2003) has gained prominence. This framework posits that agents make the best possible predictions subject to constraints on information acquisition and processing (Maćkowiak et al., 2023). As a result, individuals selectively allocate attention, prioritizing certain sources of information over others to reduce uncertainty as effectively as possible.

Returning to the analogy, learning new recipes requires time and effort that may not always be available. Instead, individuals tend to prepare satisfactory dishes using familiar ingredients and recipes they already know.

Some strands of the literature define all expectations that are not rational as “behavioral expectations.” In my view, this terminology is not ideal, because behavior refers to any action or decision made by an individual, regardless of whether it is considered rational. In this sense, the formation of expectations about economic variables is itself a form of behavior.

I therefore prefer the term subjective expectations. The Merriam-Webster dictionary defines “subjective” as “relating to or being experience or knowledge as conditioned by personal mental characteristics or states”. Accordingly, expectation formation is based on a set of experiences and knowledge shaped by how individuals perceive the world.

I would not characterize subjective expectations as inherently “worse” than rational expectations ex ante. Rather, expectation formation should be evaluated based on its consequences for individuals’ well-being, as an ex post exercise.

What do we know about subjective expectations? The standard approach to eliciting them is through surveys of households. Additionally, advances in Large Language Model (LLM) techniques allow researchers to infer these perceptions and expectations from “field evidence,” such as social media, books, newspapers, radio, and television. I refer to the first approach as an active method and the second as a passive method of eliciting expectations.

The literature has established that expectations affect both perceptions and behavior. Roth and Wohlfart (2020) expose a sample of individuals to information treatments featuring high or low forecasts of the probability of a recession. Although households exhibit more pessimistic and more dispersed priors about recession probabilities compared with professional forecasters, they update their beliefs in the direction of the information provided. The authors also document a negative effect on individuals’ subjective financial prospects, particularly regarding the perceived probability of becoming unemployed. This effect is stronger among individuals with greater exposure to past recessions, lower levels of education, lower earnings, and is more pronounced for men.

Preferences regarding changes in economic variables and policy also play an important role in shaping subjective expectations. Pfajfar and Winkler (2025) find that perceptions of the U.S. inflation target differ from the official one: respondents tend to believe that the target is 3%, rather than 2%. In addition, a non-negligible share of respondents considers deflation to be an optimal target.

Regarding preferences, most individuals report that lower inflation is either better or at least no worse for them. Preferences concerning unemployment are concentrated around an indifference point, with the modal response indicating that a reduction in unemployment would not significantly affect their well-being. Finally, preferences over interest rate changes are heterogeneous across households, with some preferring lower rates and others favoring higher ones.

When asked to choose between reducing inflation or unemployment, but not both, more than 80% of respondents prefer a reduction in inflation. More specifically, the authors estimate an acceptable sacrifice ratio for U.S. households of 0.69. This implies that, to reduce inflation by 1 percentage point, households are willing to accept an increase in the unemployment rate of 0.69 percentage points. This ratio is lower than those typically used in standard monetary policy loss functions.

Moreover, the sacrifice ratio is higher during periods of elevated inflation and among individuals with higher education and numeracy levels, as well as retirees. In contrast, it is lower for women and Latino individuals. Incorporating these findings into central banks’ loss functions would imply that optimal monetary policy places relatively greater weight on stabilizing unemployment, at least three times more, compared with standard models.

One aspect that it is interesting to analyze is the process of joint expectations, I mean, where the expectation formation of one variable does not depend only of the past realizations of the same variable, but also from others. The most common pair is the association between inflation and unemployment. There is evidence stating that households forecast higher future unemployment rates after experience a higher inflation setting. This is due to they associate supply-side mechanisms under this relationship, where they consider a higher inflation as an anticipation of worse outcomes.

Individuals perceive the impact of macroeconomic shocks in different ways. Andre et al. (2022) compare the subjective models of households with those of experts (academics and professional forecasters) in response to several hypothetical shocks, including an oil supply shock, a monetary policy shock, a government spending shock, and an income tax shock. Experts tend to interpret the effects of these shocks on inflation and unemployment in line with standard textbook models. In contrast, households often rely on different mechanisms to explain these effects, leading to predictions that differ from conventional evidence. For example, a large share of households in the survey report that, following a contractionary monetary policy shock, both inflation and unemployment increase. A similar perception is observed in response to an increase in income taxes.

The authors conclude that individuals selectively retrieve certain propagation mechanisms of shocks while neglecting others. This pattern is referred to as selective memory. In other words, individuals tend to focus on channels that are typically considered less central in standard models. In the case of monetary policy, households often associate higher interest rates with higher inflation, invoking a cost channel in which firms raise prices due to increased borrowing costs.

The paper also finds that contextual cues influence forecasts by altering the channels individuals retrieve, a mechanism referred to as associative memory.

What are the perceived links between recessions and inflation? According to the data, most past recessions have been associated with disinflation. However, Roth and Wohlfart (2020) find that households’ beliefs about the probability of a recession do not significantly affect their inflation expectations. The authors conclude that consumers may not be sufficiently sophisticated to account for complex relationships between macroeconomic variables when forming their beliefs.

One strand of the literature models subjective expectations ex ante, where a hypothesis is first formulated and then tested using data. Another strand adopts an ex post approach, seeking to uncover subjective models from observed expectations data using generalization techniques.

In this vein, Hou and Wang (2025) propose a test of joint expectation formation by extending a Kalman filter to a multivariate setting. The authors find that households associate higher inflation with subsequent episodes of high unemployment, but not the reverse. Moreover, under this pattern of expectations, joint learning generates amplified responses to standard supply shocks relative to an independent learning process, while the response to demand shocks is dampened.

In conclusion, research on macroeconomic subjective expectations is still evolving. Its development has progressed alongside advances in methods for eliciting expectations about aggregate variables. Most studies have focused on inflation and unemployment, but there is considerable scope to analyze how households perceive other variables, such as government spending, interest rates, and taxes.

There is also room to further explore the subjective processes underlying macroeconomic uncertainty. Although the literature does not provide a clear consensus on the impact of uncertainty on inflation, examining this relationship through the lens of expectations may help resolve this puzzle.

Ultimately, we need to continue uncovering the “ovens” that each household uses, as well as the “recipes” and “ingredients” that shape their expectations.

References