Herding in aid allocation

This paper investigates a claim repeatedly made, but never tested, that aid donors herd. To do so it originally uses methodologies developed in finance to measure herding on financial markets, and adapts them to aid allocation. The motivation for studying herding is to improve our understanding of aid allocation beyond observable determinants. If herding is indeed present, then it is likely to shape aid patterns in a significant way by creating aid darlings, orphans, but also by exacerbating aid volatility. Our approach starts by carefully defining aid to avoid including herding-prone aid, such as humanitarian aid and debt relief, and the sets of donors and recipients. Once this is done, herding is measured by directly applying the indexes used in finance to yearly aid data. Results show herding is indeed present, but that it is small. A second step is to introduce modifications to better match the characteristics of aid allocation. The most important in the paper is to change the time horizon. Unlike traders, aid donors commit to an aid partnership over several years, and yearly variations may contain a large part of randomness. Instead of year-to-year changes, we instead use 3 and 5-year allocations to measure herding. With this modification herding is still found to be present, but also of a larger size. It is now similar to what is traditionally found on financial markets. The next, important step is to acknowledge that aid donors' allocation decisions almost surely follow similar determinants and changes in these determinants generate a lot of co-movements. Herding measures by definition interpret these simultaneous decisions as herding, when they merely reflect common views among donors (think about a natural disaster occurring in a country that dramatically increase aid needs). Herding determinants are carefully estimated and their contributions to herding measures are then removed to obtain an estimate free of the effects of observable variables that affect aid allocation. This procedure shows that, even after taking observable factors into account, herding is still present. It suggests other considerations drive herding behavior.