Working Paper: NBER ID: w25078
Authors: Scott R. Baker; Lorenz Kueng; Steffen Meyer; Michaela Pagel
Abstract: Because of limitations in survey-based measures of household consumption, a growing literature uses an alternative measure of consumer expenditures commonly referred to as "imputed consumption." This approach typically utilizes annual snapshots of household income and wealth from administrative tax registries to calculate household spending as the residual of the household budget constraint. In this paper we use transaction-level retail investment data to assess the measurement error that can result in imputed consumption due to intra-year changes in asset values and composition. We show that substantial discrepancies between imputed and actual spending can arise due to trading costs, asset distributions, variable trade timing, and volatile asset prices between two annual snapshots. While these errors tend to be quantitatively small and centered around zero on average, we demonstrate that they vary across individuals of different types and income levels and are highly correlated with the business cycle. We end by suggesting ways to minimize the impact of these imputation errors in future research and we discuss which research questions are least likely to suffer from such errors.
Keywords: Imputed Consumption; Measurement Error; Household Spending; Administrative Data
JEL Codes: C81; D12; D14; E21; G11
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Imputation from annual portfolio snapshots (G11) | significant measurement errors (C83) |
| Imputation from annual portfolio snapshots (G11) | measurement errors correlated with household financial characteristics (C83) |
| Discrepancies between imputed and actual consumption (E21) | procyclical measurement errors (E39) |
| Households with larger portfolios (G59) | greater levels of imputation error during high market volatility (C58) |
| Measurement errors (C20) | bias empirical research outcomes (C90) |