In this note, we explain how the absence of recent data has affected our models and why, despite signs of normalization in certain dimensions of the labor market, the forecast does not converge back to the inflation target.
To what extent are the models affected by missing data?
Yes, the models have been affected by the absence of recent data. That said, most of the core models we use operate at a quarterly frequency. As such, the lack of a single month of observations—whether for labor market indicators or inflation—has only a limited impact, since we can still reconstruct or impute the quarterly values.
Naturally, if the government shutdown were to persist and compromise an entire quarter of data, the issue would become more material. For now, however, only a few models—specifically the distribution-based models and the Common-Idiosyncratic (CI) model—are meaningfully affected by the data gap.
For the CI model in particular, we are still assessing the best technical solution. The most practical approach will likely be to interpolate the missing series levels once the new data are released.
What are you doing to minimize the impact?
For the time being, the most straightforward strategy is to continue operating as effectively as possible—essentially as if the data had been released. This is why, for instance, we have circulated the new preview for the October CPI.
In the case of our machine-learning model, for example, the framework relies on more than one hundred variables, so the absence of a few published series has only a limited impact on its performance. More broadly, we are currently relying on a set of working assumptions to reconstruct missing series wherever feasible. As noted earlier, most of our models operate at a quarterly frequency, which allows us to adjust the quarterly nowcast based on these interim assumptions.
At this stage, this remains the most practical approach to mitigating the impact of missing data on our modeling framework.
Will the impact of the missing data persist in future runs?
By definition, working with an incomplete dataset inevitably affects the models. However, as noted earlier, in most cases we expect this impact to be relatively limited, since many of our frameworks operate at a quarterly frequency and will still capture sufficient information for the period.
For models that rely on monthly data, the impact will naturally be greater. Once new data become available, we will need to reassess and determine the most appropriate approach. One option, as previously discussed, is to interpolate the missing series and interpret the resulting model outputs using complementary indicators or signals from other models.
That said, it is likely that for monthly-frequency models the effects of the missing data will persist for several months, until the models have fully absorbed and adjusted to the updated information.
The labor market has normalized. Why is inflation forecasted to stay elevated?
The labor market is not the only variable feeding into our models; supply-side shocks also play a role—most notably, at present, tariff-related shocks. Yet even before the recent tariff developments, our models consistently projected inflation failing to return to target by the end of the medium-term, despite evidence of labor market normalization.
This outcome arises from different mechanisms across models. In several cases, models have begun to estimate a higher degree of inflation persistence—implying that inflation may eventually return to target, but only over a much longer horizon, extending well beyond the medium term. Other frameworks, such as the FRB model, have more recently incorporated higher inflation expectations, which mechanically feed into a higher inflation path going forward.
Moreover, as discussed in the next and last Q&A, models using alternative measures of economic slack—such as vacancy-based indicators rather than the unemployment gap—are currently producing very similar results.
What if you use V/U instead of the Ogap/Ugap in the models?
At any given time, we monitor the forecasts from our main model— which uses the employment gap as its measure of slack—against those from alternative specifications that rely on different slack indicators. At present, all versions of the model, regardless of the slack formulation, yield broadly the same conclusion: inflation is expected to moderate over time but not to return to target (at least not by the end of the medium-term). This holds true even when using labor-market variables that have already fully normalized.
As an illustration, we report below the forecast from the main model that employs vacancies—either as the vacancy rate or as the ratio of vacancies to unemployed (V/U). In both cases, the projected inflation path is very similar to that of our standard model. The same outcome was observed a year ago and also holds when the estimation ends in 2024 and conditioned on the current path of labor-market variables for 2025.
This pattern arises because, as previously noted, the labor market is not the only driver of inflation in these frameworks. Moreover, all versions of the model estimate a nontrivial degree of inflation persistence. Thus, while the labor market has indeed normalized and inflation is projected to move closer to target, it does not return to target within the medium-term horizon—even when alternative measures of slack are employed.
Figure 1. “Main” model forecast using V/U as slack
Figure 2. “Main” model forecast using the vacancy rate as slack