We show that three factors combine to explain the mean excess sensitivity reported in studies estimating the consumption response to income shocks: the use of macro data, publication bias, and liquidity constraints. When micro data are used, publication bias is corrected for, and the households under examination do not face liquidity constraints, the literature implies little evidence for excess sensitivity on average. The result holds when we control for 45 additional variables reflecting the methods employed by researchers and use Bayesian model averaging to account for model uncertainty. The estimates of excess sensitivity are also systematically affected by the size of the income shock, the choice of proxy for consumption, and the order of approximation of the Euler equation.

Reference: Tomas Havranek & Anna Sokolova (2018), "How Does Consumption Respond to Expected Income Changes? Evidence from 144 Studies." Czech National Bank and Charles University, Prague. Available at meta-analysis.cz/excess_sensitivity.