Farmers’ responses to unexpected weather variability in developing countries: The case of Indonesia

This paper aims to estimate the causal impact of the unexpected weather variation on the employment level of the farm households by exploiting natural variation of the unexpected weather changes and variation in decision of labour allocation based on gender and occupation in Indonesia. Weather variability is increasing in frequency, duration and intensity. It cannot be predicted with certainty and effectively mitigated in terms of both the time of the event and the impact of the loss (Lei, Liu, et al. 2016). Through its implications for agricultural production, the weather variability together with extreme weather will lead to crop failure, increased production costs, damaged farm infrastructure, reduced farmer incomes, and increased rural poverty (Winsemius et al. 2018). Despite substantial discussion on crop diversification, the farmers’ option to manage their family labour as means of risk avoidance is limited in the literature (Ayenew 2017). It is a crucial question to address how this strategy is effective in response to the negative effect of unexpected weather changes. We exploit the unexpected weather variability that is defined as the deviation between the real-time value of weather condition, proxies by the Standardized Precipitation and Evapotranspiration Index (SPEI). Then, we construct our model based on the assumption that farming family hold non-separable between production and consumption decision, as a response to market imperfection. To this end, our outcome variables refer to a household’s allocation of time to work (log household worked hours) of family labour to any of the categories of activities (in agriculture and non-agriculture) and by gender. While, our control variables comprise information on household and community characteristics. Household characteristics variables consist of farm or land size, parents education, household size, non-farm asset, the working-age (15-65 years old) member of the household. Village characteristics measure the availability of infrastructure that are the level of road and electricity, and irrigation. Information on altitude, experience to drought in the last year and majority income of village dwellers are also taking into account in our model. Hence, by utilising a linear household fixed-effect method, our model can be written as: y_ijt= 〖α+φD〗_jt+〖βX〗_ijt+〖ωV〗_jt+〖δ_i+δ_d+δ〗_t+ ε_ijt………(4) Overall, our results found that there were causal inferences between the employment level of farm households and weather-based variables. Unexpected variability of weather exposure reduced the number of working hours of farms employed by 4.7 per cent per standard deviation. In contrast, farm household’s member worked more in non-agricultural job as indicated by the number of working hours increased 3.6 per cent. These results are robust to the inclusion of sub-district and year fixed effect as control, and several confounding factors. Moreover, the panel regression confirmed that all policies variables have a significant positive on households working hours. Agricultural extension, public works project, and credit facilities in the villages are substantial consideration for the policies design to support farmers in overcoming the negative effect of weather variability.