What we anticipate seldom occurs; what we least expect generally happens.
■ Using Prior-Year Estimates Rather Than Revised Statistics
Historical data are based on final revised estimates rather than the estimates that were available at the time. For example, the historical levels of U.S. corn production are revised throughout the season with the final revision occurring after the end of the season. These final revised estimates for each season (the actual levels) can differ substantially from the crop estimates that prevailed during each season (the expected levels). Similarly, historical corn consumption and export levels (the actual levels based on final revised estimates) can be very different from the expected levels that prevailed during each season.
Typically, fundamental models would use actual historical data as inputs. But is this default approach the best procedure? A strong argument can be made that the data levels expected at the time are more relevant to explaining price behavior than actual data levels that only became known after the price forecast period in question. Thus, it may be possible to build a more accurate model using past estimates rather than actual statistics as the price-explanatory variables. For example, if we are trying to construct a model to explain and predict September–November corn prices, we might well find that the past production and usage estimates released during the September–November ...