Discussion paper
DP14273 Inferring Complementarity from Correlations rather than Structural Estimation
According to the Hicksian criterion, two products are complements if their (compensated)
cross-price elasticity is negative. While attractive in theory, the implementation of the
Hicksian criterion can be hard: computing elasticities requires the estimation of structural
models allowing for both complementarity and substitutability. Here, we instead
investigate the correlation criterion, whose implementation only requires the comparison
of observed market shares. We show that, in a large class of non-parametric models, the
correlation criterion satisfies all the axioms by Manzini et al. (2018) and how, in mixed
logit models, it can be used to learn about the Hicksian criterion.
£6.00
