Professional forecasters play a key role in shaping macroeconomic expectations. Their published projections influence investors, impact wage and price setting, and guide central banks. However, the process behind generating forecasts is far from frictionless. Creating a new forecast often requires updating internal models, justifying changes to supervisors or clients, and managing reputational risk. These challenges are especially significant in professional settings where forecasts are public and linked to institutional credibility. Revising forecasts too frequently can seem erratic or indecisive.
In Barcelona School of Economics Working Paper 1476, “Lumpy Forecasts,” Isaac Baley and Javier Turén examine how professional forecasters respond to real-world frictions using high-frequency data and a structural model.
Their analysis focuses on monthly US inflation forecasts from Bloomberg’s Economic Forecasts (ECFC) survey, which includes a wide range of institutions—banks, consulting firms, universities, and research centers. The data reveal that forecast revisions are infrequent and sizable when they occur (“lumpy”) and closely linked to the prevailing consensus (“strategic”). Moreover, forecasters often appear to overreact to new information. While initially puzzling, these patterns can be understood as optimal responses to the constraints of professional forecasting.
Forecasters face a tough trade-off: the desire for accuracy clashes with the need to seem consistent and aligned with peers. Frequent revisions might appear inconsistent or damage institutional credibility. Therefore, updates are made only when the incoming signal is strong enough to justify the costs—whether cognitive, reputational, or organizational. These dynamics lead to what the authors call rational inaction: a tendency to revise only when the benefits surpass the friction.
This framework sheds light on why forecast revisions may appear abrupt, exaggerated, or overly aligned and provides a more accurate lens for interpreting professional forecasts.
What the Data Tells Us: Patterns of Lumpy Revisions
The authors examined monthly inflation forecasts from the Bloomberg ECFC survey from 2010 to 2019, tracking individual predictions made by hundreds of professional forecasters. The data shows several notable patterns (see Table 1 below):
- Periods of Inaction Followed by Major Revisions
Despite monthly data releases, forecasters revise their inflation predictions only five times per year on average (an inaction rate of 58%) - Consensus Influence
Deviations from the consensus predict both the likelihood and the direction of adjustment. For instance, a forecast that is 50 basis points above the consensus is 20% more likely to revise downward, with an expected revision of –0.45 basis points - Heterogeneity Across Forecaster Types
Banks and consulting companies revise more frequently and more aggressively than academic or research institutions, which highlights differences in institutional frictions - Apparent Overreaction to New Data:
When forecasters finally revise their predictions, the revisions often exceed what the incoming data would justify
| Statistic | Value |
|---|---|
| Average forecast revisions per year | 5.1 |
| Share of zero revisions | 58% |
| Average absolute revision size | 29 basis points |
| Observations | 9,256 |
Notes: Bloomberg data, 2010-2019.
A Model of Lumpy and Strategic Forecasting
To explain these patterns, the authors develop a theoretical model where forecasters constantly update their internal beliefs using all available information but only revise their reported forecasts when doing so exceeds a cost-benefit threshold.
The model includes three main frictions:
- Learning
Beliefs update smoothly and rationally, combining new public and private information using Bayes’ law - Fixed Costs of Revision
Changing a forecast involves incurring a fixed cost, which makes frequent minor updates undesirable - Strategic Concerns
Forecasters worry about how their predictions match the consensus, so they need to forecast what the “average” forecaster will predict to stay close to it
These ingredients create an “inaction region,” which is a zone where it is optimal not to revise. Figure 1 below shows this visually:
Notes: This figure illustrates the inaction, representing the range of beliefs where forecasters prefer not to revise their forecasts. When beliefs about inflation and the consensus fall inside the shaded band, no update occurs.
How to read Figure 1:
- The red dot indicates the current forecast of 2%
- The dark band displays the range of beliefs about inflation (x-axis) or beliefs about consensus (y-axis) where no revision is optimal given the current forecast
- One blue star (inflation beliefs at 2.5% and consensus beliefs at 1.5%) is inside this region, marking where beliefs have shifted, but not enough to cause a change (forecasts stay at 2%)
- The other blue star (inflation beliefs at 1.5% and consensus beliefs at 2%) is outside the inaction region, and the forecaster resets their forecast, adjusting to a new value of 1.7%
- The slope of the inaction region reflects strategic concerns
- The width indicates the fixed cost of adjusting
- The model predicts that forecasts stay flat until beliefs shift enough to justify a jump
One key implication of the model is that forecasters’ internal beliefs about inflation shift more gradually than their published forecasts. While public forecasts might stay the same for months before a major revision, their underlying beliefs change continuously, reflecting new information as it comes in.
Figure 2 below illustrates the dynamics of beliefs and forecast behavior for one forecaster over a year in a calibrated model that accurately matches the empirical regularities.
Notes: This figure illustrates how a professional forecaster’s inflation belief (solid green line) and consensus beliefs (dashed green line) evolve continuously as new information arrives, while their reported forecasts (blue line) adjust only occasionally.
How to read Figure 2:
- The green lines represent how internal beliefs about inflation (solid lines) and beliefs about the consensus (dashed lines) evolve smoothly as new information becomes available, following the standard Bayesian learning process
- Notably, inflation beliefs converge with the actual inflation rate of 3.75%
- However, the blue line reveals that the actual forecast remains unchanged for several periods, only jumping when beliefs move outside the inaction region, and may remain far from the actual inflation realization
This figure emphasizes the main idea of Baley and Turén’s paper: forecasters are always learning, but they only update their forecasts when the benefits of accuracy exceed the costs of change.
Correcting for Strategic and Frictional Biases
These findings warn against taking forecast data at face value. Forecasts are not raw beliefs. They are processed, refined, and reputationally managed representations of underlying expectations. Recognizing this difference is essential for using survey data effectively.
To better estimate forecasters’ true beliefs, the authors suggest a two-step correction.
First, they propose active revisions, removing periods of inaction to focus on moments when forecasters respond to meaningful information.
Second, they account for strategic concerns by eliminating the influence of the consensus through regression-based residualization. This method filters out reputational pressures and provides more accurate measures of inflation expectations.
Broader Implications and Future Research
The logic behind lumpy and strategic forecasts extends well beyond professional inflation predictions. The same frictions—revision costs, strategic alignment, reputational pressure—can influence beliefs and expectations across many areas:
- Other Macroeconomic Variables
Forecasts of GDP growth, unemployment, or fiscal policy may exhibit similar lumpiness. Studying them could deepen our understanding of the dynamics of expectations in macroeconomic models. - Households and Firms
Do households and firms also revise their expectations in a lumpy way? Understanding how non-professional forecasters form and adjust their beliefs can help us improve models of consumption and investment behavior. - Macroeconomic Transmission of Shocks
Lumpy adjustments to expectations can influence how monetary policy shocks and economic news transmit through the economy. Importantly, the frictions underlying forecast inaction give rise to state-dependent behavior: during periods of high inflation volatility, forecasters revise more frequently and respond more strongly to incoming shocks.
Conclusion: Rational Revisions Under Constraints
Far from being irrational, lumpy forecast revisions represent optimal responses to a complex set of frictions. Recognizing these constraints is essential for accurately interpreting survey data, refining communication strategies, and developing macroeconomic models that more accurately reflect how expectations are formed. The message of this research is clear: expectations are influenced not only by data but also by the costs and incentives of reporting them.
