Counterfactual imputation approaches for the housing component of the October 2025 CPI

Due to the 2025 lapse in appropriations, the U.S. Bureau of Labor Statistics (BLS) could not collect Consumer Price Index (CPI) housing survey data for the October 2025 reference period. When operations resumed in November, BLS focused collection efforts on November 2025 data, and did not attempt retroactive collection.¹ This resulted in completely missing rent values for the entire month of October 2025.

BLS faced a decision about how to impute the missing rent data. It was determined the preexisting CPI hierarchical imputation algorithm would be applied, which resorts to using the last collected price for a sample unit when all other preferred imputation source pools are empty. CPI calculates price indexes for rent of primary residence (rent) and owners’ equivalent rent (OER) using 6-month collection panels that are surveyed twice a year, including a panel that is surveyed in April and again in October. So, in this unusual case of no collected rent data for an entire month, rent values from April 2025—the last time the April/October panel sample of housing units was surveyed—would be carried forward into October.² Practically, this means rent values were copied from April to October, price and index relatives were calculated, and those relatives were chained onto the September indexes. Thus, the October 2025 indexes for rent and OER were unchanged from their September 2025 values.

When BLS resumed operations following the 2025 lapse in appropriations, the agency was under a tight deadline to produce and publish indexes as soon as possible. Using the existing hierarchical imputation algorithm—which employs carry-forward imputation as the imputation method of last resort—was the only way to calculate indexes in a timely fashion.³ There are several reasons for this. First, BLS did not want to intervene and impose last-minute, untested judgment into the production process, as this could have been perceived as unscientific manipulation of the data. Second, vetting a proposed alternative imputation method and gaining endorsement takes considerable time. A methodological change of this magnitude would require advance notification with a public comment period prior to adoption. Third, BLS information technology systems were intentionally built and tested for the hierarchical imputation algorithm. The time needed to modify, test, and certify changes to those systems would have further delayed the release of the November 2025 CPI and subsequent releases.

While the decision to use carry-forward imputation was the only option available for BLS to release November CPI data as quickly as possible, no change in prices for October was an unlikely economic reality. BLS stated the effects of the carry-forward imputation method used in October 2025 would resolve in April 2026 when the April/October housing panel was collected and used again. BLS also stated that the 1-month rent change estimated for April 2026 would be atypical: it would essentially be based on the 1/6th root of a 12-month change, instead of the 1/6th root of a 6-month change.⁴ In essence, the 1-month price changes for the rent and OER indexes and for higher level aggregates would be correct for the intervening time period, although the price levels and their 12-month changes would be lower. And, when the April/October panel was collected again in April 2026, the level would be corrected.

This article explores counterfactual index results based on various alternative approaches to the carry-forward imputation methodology used for the housing component of the CPI in October 2025. The alternative approaches represent possible methods of imputing missing data that are not a part of the official CPI imputation algorithm and have not been vetted or approved for official use. The approaches, results, and analysis presented in this article are strictly for research purposes.

Technical background

Before exploring alternative imputation approaches, a further explanation of the housing survey and the unique methodology used to sample housing units and calculate changes in rents is necessary. To administer the housing survey, housing units are selected for sampling from small subsets of each larger geographic area represented in the CPI geographic sample.⁵ Within these subsets, or segments, the sampled housing units are further divided into six mutually exclusive panels. Each panel is surveyed twice a year. Panel 1 is surveyed in January and July, Panel 2 is surveyed in February and August, and so on.

BLS quality adjusts collected rents with an age-bias adjustment to account for the physical depreciation of a housing unit over time.⁶ Housing unit-level and neighborhood characteristics are used to estimate quality-adjustment values. Adjustments are applied equally to all units in the same CPI sampling area. The 6-month rent relative for a housing unit used in the rent or OER indexes is calculated as the rent in the current month divided by the rent from 6 months ago plus the age-bias adjustment. However, age-bias adjustments are only made to housing units where BLS collected the rent data in the current period; these units are then used as potential sources for the imputation of units where BLS was unable to collect the rent data. For October 2025 data, when every unit was imputed via carry-forward methodology, no age-bias adjustments were applied to the April 2025 rents, and the October 2025 indexes were unchanged from September 2025.

The rent estimator uses the change in the economic rent (the contract rent adjusted for any changes in the quality of the housing unit) to estimate the change in the average rent. The sum of the current period economic rents for each usable housing unit within a segment, weighted by the renter weight for that segment, is divided by the sum of the weighted economic rents from 6 months earlier. This ratio represents the 6-month change in rent for all renter-occupied units within a segment.

The OER estimator uses the change in the pure rent, which excludes the cost of any utilities included in the contract rent. In a parallel calculation to the rent estimator, the sum of the current period pure rents for sampled, renter-occupied units within a segment, weighted by the owner weights, is divided by the sum of the weighted pure rents from 6 months earlier. This ratio represents the 6-month change in OER for all owner-occupied units within a segment.

The rent and OER indexes are derived by applying the sixth root of the 6-month rent change to the index value from the previous month.⁷ This derivation converts the 6-month rent change into an average monthly change over the 6-month interval between sampling periods for each panel.

Counterfactual approaches

Four alternative approaches for the imputation of missing rent data for October 2025 were defined for this article, described below. BLS used each approach to calculate the rent and OER research indexes for each of the 32 basic geographic areas. Upper-level aggregation to the U.S.-level index was then applied in each approach.

1. Backcast: This approach uses the behavior of another panel as a proxy for the missing panel by substituting the November 2025 index change for the October 2025 index change. While this approach is simple to compute and explain, it assumes October prices moved exactly like November prices.⁸

2. Interpolate: This approach uses the adjacent period indexes as bounds for the missing index by geometrically averaging the September 2025 and November 2025 index changes. These average changes are then substituted for the October 2025 index changes. While this approach is simple to compute and explain, it is not timely since November data is needed to determine October.

3. Forecast: This approach uses the historical trend of the April/October panel to forecast the missing index by geometrically averaging the 6-month index changes from April 2021 forward. These average changes are then substituted for the October 2025 index changes. This approach is relatively simple to compute, timely, and benefits from being informed by historical data from the same panel. However, October could have been an off-trend month.

4. Categorical average: This approach uses historical data as the basis for imputing rent change at the unit level. Using data from April 2021 forward for only the April/October sample of housing units, this approach determines at a housing unit level whether the rent increased, decreased, or was unchanged for each month. Then, for each of the 32 basic areas, the percentage of units that increased, decreased, or were unchanged is determined.⁹ Next, each unit in the October 2025 sample is randomly assigned to increase, decrease, or not change based on historical data for that unit, and the average value is applied. This approach can be done in two different ways: including age-bias adjustments and excluding age-bias adjustments. This approach is timely, informed by historical data, and makes it possible to calculate variances, but is difficult to compute and explain.

Results

The CPI program used the four counterfactual imputation approaches to calculate basic area indexes for both rent and OER and then aggregated these basic area indexes to the U.S. level using the standard CPI Laspeyres formula. The results were then compared with the published values of these indexes for the September 2025 through April 2026 period. (For the purposes of this research, the October 2025 CPI value shown matches the published CPI value for September 2025 throughout this article.) The expectation was that regardless of approach, the rent and OER indexes would return to the same level in April 2026, although this would depend somewhat on the number of units that were collected with rents in April. The return would be delayed if many units were not collected. Further routine sample rotation, application of the age-bias adjustments, and updated weights would impact the ability of the indexes to return to precisely the same level.

Charts 1 and 2 illustrate the published CPI for rent and OER, respectively, and the research indexes based on the four counterfactual approaches, from September 2025 to April 2026 at the U.S. city level. The research indexes deviate from the published rent and OER indexes in October 2025 when the April/October panel was not collected, leaving the published indexes unchanged from their September values.

The backcast and interpolate approaches rely on index data from periods adjacent to October 2025 when rent and OER inflation were relatively low and, thus, those indexes diverge the least from the published indexes. The forecast and categorical average approaches rely on historical data from periods when rent and OER inflation were relatively high and, thus, those indexes diverge the most from the published indexes. All the indexes exhibit the same trends between November 2025 and March 2026 (when the other five housing panels were collected), with the published rent and OER indexes running below the research indexes. Then, in April 2026, the published and research indexes converge to the same relative values (between 445.021 and 445.101 for rent and between 439.082 and 439.115 for OER) once the April/October panel is collected again.

Tables 1 and 2 display the differences in 1-month percent changes between the published CPI for rent and OER, respectively, and the research indexes based on the counterfactual approaches. These figures quantify the index behavior described previously, where the published and research indexes deviate in October 2025, move in tandem between November 2025 and March 2026, and converge in April 2026.

Charts 3 and 4 illustrate the 12-month percent changes for the rent and OER indexes, respectively, and for the research indexes based on the counterfactual approaches from September 2025 to April 2026 at the U.S. city level.

The 12-month percent changes differ among the indexes starting in October 2025, with the published indexes running below the trend of the research indexes. It is only with the April 2026 indexes, and the collection of the April/October panel of housing units, that the trends for the published and research indexes converge to between 2.79 and 2.81 for rent and between 3.28 and 3.29 for OER.

Tables 3 and 4 display the differences in 12-month percent changes between the published CPI for rent and OER, respectively, and the research indexes based on the counterfactual approaches. These figures quantify the behavior of the 12-month trends described previously, where the published and research indexes deviate in October 2025, move in tandem between November 2025 and March 2026, and converge in April 2026.

Conclusion

The experiments conducted and described in this article demonstrate that regardless of which counterfactual approach is applied to impute missing data in October 2025, the research indexes and the published indexes for rent and OER all returned to the same general index level in April 2026. The small differences between the levels of the published indexes and research indexes are attributed to routine sample rotation, the application of age bias adjustments, and updated unit-level weights. This article also validates the BLS conclusion that the effects of the carry-forward imputation method used in October 2025—namely the reduction in index levels and 12-month changes for the rent and OER indexes—would resolve in April 2026 once the April/October panel was collected again. In the future, BLS may consider incorporating improved exception-case methodology to better position CPI business operations and index calculation methodologies to address large-scale data interruptions in a timely manner.