Incorporating microdata into macro models
By Stuart McIntyre
Macroeconomics is undergoing an important shift. Economists now have access to unprecedentedly rich microdata on households and firms. However, many macroeconomic models looking at the economy as a whole still rely primarily on indicators such as GDP, inflation and unemployment. This disconnect has prompted growing academic interest in frameworks that can better integrate detailed microdata into macroeconomic analysis.
In a new paper, Incorporating Microdata into Macro Models Using Pseudo-VARs, we propose a method for embedding very detailed microdata directly into macroeconomic time-series models. The aim is simple but important for economic policymaking: to understand not only how the economy responds to shocks, but also who within the economy is affected.
Macroeconomic analysis traditionally compresses vast amounts of information into a small number of summary statistics. For example, analysts may use survey data to derive indices of income inequality or to split people into broad income groups. While useful, these indicators hide much of the variation across individual households or businesses in the survey.
This matters for policy because economic shocks rarely affect everyone equally. Recessions may hit lower-income households harder, while changes in interest rates can have very different effects on borrowers and savers. To capture these differences, models need to look inside the data rather than just relying on summary averages.
However, incorporating microdata raises significant econometric challenges. Microdata often takes the form of cross-sectional snapshots of different individuals or businesses at various points in time, rather than tracking the same people and firms over time. Incorporating thousands of individual observations into time-series models can quickly become computationally and statistically infeasible.
Vector autoregressions (VARs) are a core tool in empirical macroeconomics. They model how multiple variables evolve jointly over time, using past values of each variable to explain current outcomes. VARs are attractive because they require relatively few structural assumptions and can capture dynamic interactions across macroeconomic indicators such as output, inflation and interest rates. This flexibility has made them a workhorse for analysing business cycles and policy shocks.
Standard VARs, however, use broad, summary data and are not designed to incorporate large volumes of microdata directly. Earlier attempts to bring distributional information into VAR analysis have typically relied on compressing microdata into a small number of summary measures, like inequality scores or income group statistics, and then including these in the model. While practical, this approach inevitably sacrifices much of the richness of the underlying data.
Recent methodological work has explored ways to move beyond this limitation through so-called functional VARs. Instead of modelling a small set of individual variables, functional VARs treat entire distributions as objects that evolve over time. For example, rather than including a single measure of income inequality, a functional VAR might incorporate the entire spread of incomes and model how that distribution responds dynamically to macroeconomic shocks. In this framework, each period’s distribution is represented as a function or curve, and the VAR captures how these functions/the curve change over time.
Our paper proposes a related but more directly implementable route. Rather than compressing microdata into a few indicators, we develop a pseudo-VAR framework that embeds micro-observations alongside macro variables through the construction of pseudo-individuals. The ‘pseudo-individuals’ are simulated records that act like real people, letting the model capture detailed patterns without needing every single person’s data. These pseudo-individuals keep the important details of the microdata, while still letting us use standard VAR methods to analyse the system. This makes it possible to incorporate rich microdata into a manageable macroeconomic time-series framework without collapsing the data into a handful of summary measures.
Our pseudo-VAR approach allows researchers to identify which groups of households or businesses are most sensitive to particular shocks. Traditional average effects can be studied at the same time, improving the analysis of economic dynamics and the evaluation of policy interventions.
Crucially, the framework is designed to work with the type of microdata most commonly available to statistical agencies and researchers: repeated cross-sectional surveys. As a result, the approach can be applied in a wide range of real-world settings without requiring new data collection, making it a practical tool for integrating distributional information into macroeconomic analysis.
To illustrate the approach, we apply the pseudo-VAR framework to earnings data for the USA to study how business cycle shocks affect the distribution of labour income. The results show clear distributional effects over the cycle. When the economy is growing, people at the lower end of the earnings scale tend to see their pay rise faster than the average, while those at the top end gain less. This means that during these upswings, income differences shrink temporarily.
A key advantage of the pseudo-VAR approach is that it allows us to go beyond distributional summaries and identify who is most affected. We find that the individuals most sensitive to the business cycle are men in the lower deciles of the earnings distribution (see figure below), those with less than a college education in the lower percentiles (see discussion paper), and younger workers (see discussion paper). This highlights how macroeconomic fluctuations can have markedly different effects across groups within the labour market.
Figure 1 below shows the green (red) areas where the effect of the shock is statistically significantly positive (negative) on each decile of the income distribution (y axis) and over time periods (x-axis) for men and women. In the right-hand panel we highlight in green/red only those cases where the difference between men and women is statistically significantly different from zero.
Figure 1: Distributional IRFs for men and women
Policymakers are increasingly concerned not just with aggregate stability, but with distributional consequences across income groups, regions and types of firms. Models that combine macroeconomic dynamics with detailed microdata can provide a stronger real-world basis for assessing these trade-offs and anticipating who is most affected by economic change.
As the availability of detailed microdata continues to expand, methods that connect micro behaviour to macroeconomic dynamics will become ever more important. By offering a practical framework for doing so, this paper helps bridge the gap between studying individual-level distributional effects and looking at the economy as a whole.
Macroeconomics is shifting as researchers gain access to rich data on households and firms, but many models still rely on broad indicators like GDP and inflation, hiding important differences across individuals. A new ESCoE paper introduces a framework that embeds detailed microdata into macro time-series models. Using simulated “pseudo-individuals,” it preserves key patterns while allowing standard VAR methods, making it possible to see how shocks affect different groups and connect micro-level and macroeconomic analysis.