Three EconMacro Posts Readers Consulted Most Over the Last Three Months

Over the last three months, three EconMacro posts have attracted particularly sustained attention. They are different in format: one is a methodological note on local projections and VARs, one is a data-oriented post on United Nations General Assembly ideal points, and one is a Stata implementation guide for state-dependent local projections using xtlp. Yet they all speak to the same broader concern in applied international macroeconomics: how to move carefully from data to identification, from identification to estimation, and from estimation to interpretation.

The three posts are:

Taken together, these posts form a compact reading list for researchers interested in geopolitical risk, international political economy, panel local projections, and applied macroeconometrics with Stata.

1. Local projections are flexible estimators, not identification devices

The first post clarifies a point that is often misunderstood in empirical macroeconomics: local projections are not, by themselves, a solution to the identification problem. They are a flexible way to estimate dynamic responses once the shock has already been identified.

This distinction matters because local projections are now widely used in applied macroeconomics. They allow the researcher to estimate impulse responses horizon by horizon, rather than imposing the complete dynamic structure of a vector autoregression. This flexibility is useful, especially when the VAR propagation mechanism may be too restrictive. But flexibility is not the same as exogeneity.

The post explains this point through the example of geopolitical risk, inflation, and GDP growth. In a recursive VAR with geopolitical risk ordered first, the identifying assumption is that the innovation to geopolitical risk is not contemporaneously caused by inflation or GDP-growth innovations. The corresponding local projection must use the same residualized geopolitical-risk innovation if the comparison between VARs and LPs is to be meaningful.

The main lesson is simple: local projections do not make shocks exogenous. They make the estimation of dynamic responses more flexible once the shock has been credibly identified.

This is why the post is useful for researchers writing empirical papers. It provides a precise way to describe what a VAR-LP comparison is doing. Such a comparison should not be presented as a comparison between two identification strategies. It is better described as a robustness check on the dynamic propagation mechanism, conditional on the same identifying assumption.

The post is particularly relevant for annual geopolitical-risk applications. Annual data are convenient for cross-country macroeconomic analysis, but a year is a long period. Many channels can operate within one year: energy prices, public debt, military spending, exchange rates, shortages, confidence effects, and trade disruptions. In such settings, local projections are useful because they reduce dependence on a tightly parameterized VAR law of motion. But the credibility of the causal interpretation still rests on the identifying assumption.

Read the post here: Equivalence of Cholesky identified shocks in VARs and Local Projections.

2. UNGA ideal points: geopolitical alignment as data

The second post focuses on data rather than estimation. It presents new data on United Nations General Assembly ideal points, based on voting behavior in the UNGA. These data are central for researchers who want to measure geopolitical alignment, foreign-policy proximity, or shifts in international political preferences over time.

The post highlights an important practical feature of the updated data: the ideal-point estimates are now based on years rather than UNGA sessions. This is useful because many empirical researchers organize macroeconomic, financial, and political-economy datasets at the annual frequency. Year-based ideal points are therefore easier to merge with country-year panels on growth, inflation, reserves, debt, capital flows, geopolitical risk, or institutional variables.

This is more than a technical convenience. In empirical international macroeconomics, measurement choices often determine what can be studied. If geopolitical alignment is measured at a frequency or unit that does not correspond to the rest of the data, the researcher must either transform the variable or accept a mismatch between political and macroeconomic timing. A year-based measure reduces that friction.

The post also explains that the dataset contains different types of ideal-point estimates, including estimates based on final-passage votes and estimates based on all votes. This distinction matters because broader voting information may be more precise, but it can also reflect years in which particular issues dominate the voting agenda. Researchers therefore need to think carefully about which measure is best suited to their empirical question.

The broader lesson is that geopolitical variables are not just controls. They are constructed measures, and their construction affects interpretation.

For applied work on geopolitical fragmentation, sanctions, strategic alliances, reserve accumulation, trade exposure, or political distance, UNGA ideal points offer a transparent and reproducible measure of international alignment. The post is therefore useful not only as a data announcement, but also as a reminder that political-economy variables require the same care as macroeconomic variables.

Read the post here: New Data on United Nations General Assembly Ideal Points.

3. Testing state dependence with `xtlp`

The third post is a practical Stata guide. It explains how to test whether two impulse responses are statistically different across regimes when estimating state-dependent local projections with the xtlp package.

This question arises frequently in applied economics. Researchers often estimate separate responses depending on whether an economy is in a high- or low-risk state, a high- or low-debt state, a high- or low-inflation state, or a high- or low-institutional-quality state. The resulting impulse-response functions may look different. But visual differences are not enough. The relevant econometric question is whether the regime-specific coefficients are statistically different at a given horizon.

The post shows how to split a shock into two regime-specific variables. In the example, a vulnerability shock is interacted with two regime indicators based on religious tensions:

gen double sh_D3_1 = D3   * Dvul100
gen double sh_D3_0 = D3_0 * Dvul100

The local projection is then estimated with both shocks included jointly. The key test is:

test (_b[sh_D3_1] = _b[sh_D3_0])

The implementation detail is that xtlp stores coefficient vectors and variance-covariance matrices horizon by horizon. Therefore, the post proposes a small helper program that posts each horizon-specific coefficient vector and variance-covariance matrix as the active estimation result before applying Stata’s test command.

This makes it possible to run a Wald test at each horizon and move from an informal statement such as:

The two impulse responses look different.

to a formal statement such as:

The two impulse responses are statistically different at horizon h.

The post is also useful because it emphasizes specification comparability. Results obtained with xtlp and locproj will be directly comparable only if the estimation methods and controls are aligned. For example, method(spj) uses a split-panel jackknife estimator, whereas a standard fixed-effects specification in another command may not. The statistical test may be conceptually the same, but the numerical results correspond to the specification actually estimated.

Read the post here: Statistical Difference between Regimes in State-Dependent Local Projections using XTLP.

A suggested reading order

For readers interested in applied macroeconometrics, I would suggest the following order.

Start with the post on Cholesky-identified shocks in VARs and local projections . It clarifies the identification issue.
Then read the post on UNGA ideal points . It illustrates how geopolitical alignment can be measured in country-year datasets.
Finally, read the post on state-dependent local projections with xtlp . It provides the Stata implementation needed to test whether dynamic responses differ across regimes.

In short, these three posts cover three essential steps in empirical research: identification, measurement, and inference. This may explain why they have been among the most consulted posts on EconMacro over the last three months. They address practical problems that many researchers face when working with geopolitical variables, panel data, and local projections.