Abstract
The means of daily maximum temperatures using twenty years of record are obtained. The average of these means over n consecutive calendar days is used as a predictor of the daily maximum temperature one and more years in advance. Data from ten stations in the United States for the period 1905 to 1957 are analyzed in this fashion and a series of predictions made for several values of n. If singularities are sufficiently large and persistent, then the averages over only a few days (small n) should be better predictors than averages over a larger number of days. It is found that the prediction errors decrease with increasing n and it is concluded that, whenever averages are used to estimate daily temperature values far in advance, any singularities, even if they exist at all, are much too small to be useful.
