Climate variability, associated with farm-income variability, is recognized as one of the main drivers of livelihood diversification strategies in developing countries. Analysing determinants of livelihood diversification choices, to better understand household strategic behaviour in the event of climatic risks and other shocks, is important for the formulation of development policies in developing countries highly dependent on rain-fed agriculture, like Malawi. We use geo-referenced farm-household-level data collected in 2010-11 to investigate how climatic variability influences the pattern of diversification that farmers adopt, and the impacts of these choices on welfare. To do so we apply the 'vulnerability to expected poverty' approach which measures the future level of poverty taking into consideration the role of risk and uncertainty. The analysis considers the effect of policies and institutions such as fertilizer subsidies, extension services, safety-net and credit on diversification choices. The results show that higher levels of climate risk generally increase the likelihood of diversification across labour, cropland and income, suggesting the importance of diversification as a response to constraints imposed by increased risk. In contrast, we find that in areas with favourable average rainfall conditions households are more likely to diversify income, suggesting diversification as a response to opportunities. In terms of welfare, the analysis performed on the components of vulnerability to poverty provides evidence that climatic variables are key determinants of both components of vulnerability (expected consumption and its variance). Fertilizer subsidies are found to be significant in diversification choices for all dimensions and also particularly effective in reducing vulnerability to poverty in high variability environments although the same does not hold for extension. Looking at differences across gender, we find that women labour diversification is less responsive than men's, resulting in a lower positive impact on expected consumption per capita.