Naïve Forecasting

Naïve Forecasting is the simplest type of time series forecasting algorithm. It chooses the most recent value as the next prediction. Using the ice cream sales example, if yesterday’s ice cream sale was $960, the Naïve Forecasting algorithm will predict today’s ice cream sale as $960. This algorithm is used when the time series data is quite random and does not include any seasonality or other clear patterns; in other words, the forecastability of the time series is low.

Forecastability of a time series refers to the ease with which its future values can be predicted. For example, in a time series comprised of coin tosses, with the value 1 assigned for heads and 0 assigned for tails, a random sequence is generated and the forecastability of the realized series over time is very low (extremely difficult to forecast). On the other hand, if a time series is following a clear sinusoidal pattern, then it becomes much easier to predict its future values. In cases where the forecastability of a time series is high, use of Naïve Forecasting should be avoided.

This algorithm is most effective when:

• A very small number of data points are included in the time series (more sophisticated local algorithms won’t perform well).

• The data is relatively random, no clear seasonality or trend exists, and the forecastability is low(difficult to forecast).

Naïve forecasting generally is used for benchmarking purposes. It helps you to understand how well a sophisticated forecasting algorithm is performing when compared to Naïve forecasting.