Measuring Economic Growth with a Fully Identified Three-Signal Model
Andrea Civelli, Arya Gaduh, and Ahmed Sadek Yousuf (2024). The Review of Economics and Statistics, advance online publication.
Journal article and DOI · NBER Working Paper 31517 · Title links to the NBER manuscript version.
We extend a two-signal model of economic growth by introducing a third signal that identifies the model parameters under the stated measurement assumptions. The framework recovers the optimal weight on official GDP in a composite growth estimate without calibrating GDP’s signal-to-noise ratio. An application using nighttime lights and urban land cover examines how that weight varies with the quality of national statistics.
Ahmed Sadek Yousuf (2026). Working paper.
This paper develops GMM estimation framework for bivariate spatial panel structural VARs with cluster-specific interactive effects. When cluster sizes remain fixed, factor-score estimation error induces within-cluster residual covariance. I characterize the quadratic weighting matrices that eliminate contributions from arbitrary such covariance and combine them with factor estimation from the second structural equation to preserve the recursive moment. The paper establishes consistency and inference for fixed clusters as the time dimension grows; joint cluster–time inference requires additional uniform-expansion and network-dependence conditions and allows a nonzero centering term. Monte Carlo experiments examine estimation and two-stage inference. An application to development-finance commitments and nighttime lights in African regions estimates a positive contemporaneous relationship, with light responses diminishing at later horizons.
Ahmed Sadek Yousuf (2026). Working paper.
This paper examines how the choice of higher-order cumulant restrictions affects identification and estimation in the triangular model of Lewbel, Schennach, and Zhang. I derive rank conditions showing when two fourth-order restrictions, or an augmented third- and fourth-order selection, identify coefficients even when the original selection loses rank. Identification uses the model’s latent-disturbance independence and a maintained sign restriction to label the structural coefficient. I compare joint GMM with variance-constrained concentrated GMM and derive cumulant adjustments for fixed effects and unequal country sizes. Monte Carlo results favor the fourth-order alternatives in several rank-deficient or nearly deficient designs. Applications to hartals and satellite-observed nitrogen dioxide in Bangladesh, and conflict and nighttime luminosity in Africa, illustrate their empirical implementation.
Ahmed Sadek Yousuf (2025). Working paper.
This paper develops direct Poisson estimation of a spatial autoregressive exponential conditional mean with individual and common time effects. Concentrating the spatially transformed individual effects and profiling time effects reduces numerical optimization to the structural parameters; analytical profile scores facilitate estimation while retaining zero counts. Under predetermination, feedback from count innovations to future regressors motivates temporal split-panel jackknife correction. The correction’s first-order justification uses a common expansion across the full and half panels. Monte Carlo experiments show substantial improvements in slope estimation under predetermination, with smaller gains for the spatial parameter. An application to monthly Pittsburgh crime counts finds positive spatial dependence and a positive conditional association between lagged Part II and Part I crimes, with estimates sensitive to common time controls.
Higher-Order Cumulant Identification under Common Confounding.
Nonclassical Measurement Error in Satellite-Based Economic Measurement.
Daily Night Lights and Conflict in Bangladesh} : (With Dr. Christophe Muller at Aix Marseille School of Economics): Leveraging daily night lights data (sourced from NASA), alongside Dr. Christophe Muller, we investigated how political violence (and its various definitions, such as hartals) affect (daily) economic activity in Bangladesh at (sub-district level). Here we are looking to identify short term effects of for example hartals (in the context of Dhaka, it may be seen as a dip in regular activity just after a political violence / hartal incidence and therefore very likely to be picked up by night lights). Given that the daily night luminosity data is recorded at 1:30 am local time, thus the main explanatory variable (conflict event count) is lagged by one day. The methodological challenges came from accommodating the non stationary nature of such high frequency night lights data which would have made ordinary inferences (using conventional inference tools like Ordinary Least Squares) quite difficult. Thus time series ARCH (Autoregressive Conditional Heteroskedasticity) models were used to model these interactions (between night lights and political violence). Political Violence data is derived from ACLED (Armed Conflict Location and Event Data project). From the dataset, we use keywords to further derive statistics for different definitions of violent incident (such as hartal), not otherwise included in the ACLED description. As such, most prominently, we already find a statistically significant effect of -0.9 percent immediate effect of hartals (aka political strikes) on daily night lights in Dhaka city for example. This study differs in that in the conflict literature by and far yearly data was used, and this is the first instance of using such high frequency data to identify immediate/ short term effects.