Forecasting with growth curves: The effect of error structure

The logistic model is used for forecasting a number of different types of variable; adoption of durable goods, consumption, growth of human populations. This study investigates the effect of the error structure assumed, on the forecasting accuracy of the logistic. A local logistic model is developed with an additive error term, the variance of which can be made a function of the level of the variable being modelled and its distance from saturation. Evidence is found that suggests that the variance of the disturbance term, when using the logistic to forecast human populations, is proportional to at least the square of population size. Results for the adoption of durable goods are mixed. An error structure consistent with the intuitively reasonable binomial model performed well for some series but not for others. This suggests that some data sets may be inconsistent with the binomial model.