The relative risk aversion (RRA) coefficient of a household whose consumption is close to the subsistence level may be very high. For example, if consumption is exactly at the subsistence level, the household may not be willing to bear any risk. If this is the case, then the RRA coefficient must be a decreasing function of wealth for poor households. Therefore we should allow the possibility of decreasing RRA (DRRA) in testing the full risk sharing hypothesis. However, existing tests in the empirical literature are derived using preferences that exhibit either increasing or constant RRA even when they are applied to data containing low-income households. We therefore use a Hyperbolic Absolute Risk Aversion (HARA) utility, which implies increasing, constant, and decreasing RRA as special cases, to test full risk-sharing hypothesis. Using the International Food Policy Research Institute (IFPRI) and the International Crops Research Institute of the Semi-Arid Tropics (ICRISAT) household level data sets, we find evidence in support of the DRRA hypothesis, along with evidence favoring full risk-sharing hypothesis at the village level, and evidence against the hypothesis at the inter-village level. When RRA is restricted to be constant, we replicate the previous results in the literature: reject the full risk-sharing hypothesis at both levels. Our tests, however, reject this restriction and favor DRRA in almost all cases. These results suggest that it is important to allow for DRRA in testing the full risk-sharing hypothesis when data containing low-income households are investigated.