Quantile crossing has been a challenge for quantile regression, leading to research in how to obtain monotonically increasing quantile estimates. While important contributions, these papers do not provide insight into how enforcing monotonicity influences the estimated coefficients. This paper fills this gap and shows that non-crossing constraints are a type of fused-shrinkage. The proposed estimator has good fit and (fused) variable selection properties: it can reliably identify quantile varying parameters. We investigate the 'heat-or-eat' dilemma and show that prepayment has a non-linear impact on households' consumption choices. In a growth-at-risk application the estimator has the best forecast performance.