'Program numbers' from a sample of IMF-supported programs are studied as if they were forecasts, through statistical analyses of the relationship between projections and outcomes for growth, inflation, and three balance of payments concepts. Statistical bias is found only for projections of inflation and official reserves. Statistical efficiency can be rejected for all variables except growth, suggesting that some program projections were less accurate than they might have been. Nevertheless, most projections are found to have some predictive value. Since several findings are shown to be sample-dependent, the full-sample results should be interpreted cautiously.