Modelling stock-recruitment relationships to examine stock management policies

Simulation studies are used widely for fish stock management. In such studies, forecasting future recruitment, which can vary greatly between years, has become an essential part of evaluating management strategies. We propose a new forecasting algorithm to predict recruitment for short- or medium-term stochastic projections, using a stock- recruitment relationship. We address cases in which the spawning stock has dropped below previously observed levels, or in which predicted recruitment is situated close to the maximum observed level. The relative prediction error of seven existing algorithms was compared with that of the new model using leave-one-out cross-validation for 61 data sets from ICES, the Japanese Fisheries Agency, and PICES. The new algorithm had the smallest prediction error for 49 of the data sets, but was slightly biased by the precautionary treatment of predictions of high recruitment.