EVIDENCE HAS BEEN PRESENTED that contradicts tbe current financial neural network development heuristic, wbich implies that greater quantities of training data necessarily produce better-quality forecasting models. A new time series model effect, termed tbe Time-Series Recency Effect, has been demonstrated to work consistently across neural network models for six different currency excbange time series. Tbe TS Recency Effect claims that model building data tbat is nearer in time to the out-of-sample values to be forecast produces more accurate forecasting models.