Search results

Abstract
The long-recognised spurious regression problem can lead to mistaken inference in panel instrumental variable estimation. Spurious correlations arising from correlated cycles in finite time horizons can make irrelevant instruments appear strong with signable consequences for estimated instrumental variable coefficients, or interfere with valid inference of causal effects from instrumental variable coefficients estimated using relevant instruments. The inclusion of time fixed effects in interacted specifications does not always resolve these problems. We demonstrate these concerns by revisiting recent studies of the causal origins of conflict. We offer diagnostic and corrective recommendations for avoiding the pitfalls arising from time series exhibiting persistence.

We consider an artificial population of forward looking heterogeneous agents making decisions between schooling, employment, employment with training and household production, according to a behavioral model calibrated to a large set of stylized facts. Some of these agents are subject to policy interventions (a higher education subsidy) that vary according to their generosity. We evaluate the capacity of Instrumental Variable (IV) methods to recover the population Local Average Treatment Effect (LATE) and analyze the economic implications of using a strong instrument within a dynamic economic model. We also examine the performances of two sampling designs that may be used to improve classical linear IV; a Regression-Discontinuity (RD) design and an age-based sampling design targeting early career wages. Finally, we investigate the capacity of IV to estimate alternative causal parameters. The failure of classical linear IV is spectacular. IV fails to recover the true LATE, even in the static version of the model. In some cases, the estimates lie outside the support of the population distribution of returns to schooling and are nearly twice as large as the population LATE. The trade-off between the statistical power of the instrument and dynamic self-selection caused by the policy shock implies that access to a strong instrument is not necessarily desirable. There appears to be no obvious realistic sampling design that can guarantee IV accuracy. Finally, IV also fails to estimate the reduced-form marginal effect of schooling on wages of those affected by the experiment. Within a dynamic setting, IV is deprived of any causal substance.

In dynamic panel regression, when the variance ratio of individual effects to disturbance is large, the system‐GMM estimator will have large asymptotic variance and poor finite sample performance. To deal with this variance ratio problem, we propose a residual‐based instrumental variables (RIV) estimator, which uses the residual from regressing Δyi,t−1 on as the instrument for the level equation. The RIV estimator proposed is consistent and asymptotically normal under general assumptions. More importantly, its asymptotic variance is almost unaffected by the variance ratio of individual effects to disturbance. Monte Carlo simulations show that the RIV estimator has better finite sample performance compared to alternative estimators. The RIV estimator generates less finite sample bias than difference‐GMM, system‐GMM, collapsing‐GMM and Level‐IV estimators in most cases. Under RIV estimation, the variance ratio problem is well controlled, and the empirical distribution of its t‐statistic is similar to the standard normal distribution for moderate sample sizes.

We investigate if, and under which conditions, the distinction between dictatorial and incentive-based policy interventions affects the capacity of Instrument Variable (IV) methods to estimate the relevant treatment effect parameter of an outcome equation. The analysis is set in a non-trivial framework, in which the right-hand side variable of interest is affected by selectivity, and the error term is driven by a sequence of unobserved life-cycle endogenous choices. We show that, for a wide class of outcome equations, incentive-based policies may be designed so to generate a sufficient degree of post-intervention randomization (a lesser degree of selection on individual endowments among the sub-population affected). This helps the instrument to fulfill the orthogonality condition. However, for a same class of outcome equation, dictatorial policies that enforce minimum consumption cannot meet this condition. We illustrate these concepts within a calibrated dynamic life cycle model of human capital accumulation, and focus on the estimation of the returns to schooling using instruments generated from mandatory schooling reforms and education subsidies. We show how the nature of the skill accumulation process (substitutability vs complementarity) may play a fundamental role in interpreting IV estimates of the returns to schooling.