Investing in land to change your risk exposure? Land transactions and inequality in a landslide prone region

•Conceptual and empirical analysis of the evolution of landholdings and exposure to risk over a farmer’s lifetime.•Georeferenced panel dataset of land transactions.•Evidence for disaster avoidance behaviour in the presence of landslide risk in Uganda.•Land transactions reduce but do not eliminate inequality in exposure to landslides.•Unequal exposure to landslides and inequality in initial land holdings propagate inequality in future land holdings. The poor and vulnerable tend to be increasingly exposed to natural hazards such as landslides. Land markets are one of the channels through which farmers get exposed to such hazards. This paper investigates the consequences of land transactions for the (un)equal distribution of exposure to landslide risk and of total land holdings in a rural area in Western Uganda. We propose and empirically test a mechanism through which land holdings and exposure to landslide risk evolve over a farmer’s lifetime. A structured household survey and detailed information on land transaction as well as georeferenced information on plots was used to construct a panel dataset of land transactions. Regressions with household fixed effects were run to identify how landholdings and exposure to landslide susceptibility evolves over a farmer’s lifetime. We find that farmers that are initially more exposed to landslides manage to reduce their average exposure to some extent by acquiring plots outside landslide prone areas. This goes at a cost, as farmers that are initially highly exposed acquire land more slowly than farmers that have a lower exposure on their first plot. Over a lifetime, in our case study, land transactions therefore have a somewhat levelling effect on inequality in exposure to landslides, but increase the inequality in land ownership. As such, one of the ways through which unequal risk exposure contributes to propagating inequality in total land ownings is theoretical and empirically identified.