In Part 1 of this series of articles, we discussed moving averages as a basic forecast method. The idea behind moving averages is that recent history is a good indicator of current values. For example, if a call center had 1000, 1100 and 900 calls each month for the past three months, then 1000, an average of the past three months, is a reasonable estimate for the current month. Moving averages have the advantages of being simple and flexible. We can easily vary the number of time periods we consider when forecasting a measure. One drawback of this calculation is that it considers each of the previous time periods equally.

Consider a durable goods manufacturer that maintains a warranty claims fund to pay the cost of servicing appliances under warranty. The company will want to hold a sufficient reserve to cover future costs without tying up extra cash. A moving average can be used to estimate the funds required for the next quarter as show in Table 1.

We can generally assume an average of recent history is a good predictor of near current performance if major influences on the outcome remain the same. If the manufacturer is not introducing new models, changing warranty policies, improving quality controls or making other changes then moving averages would be appropriate. One problem with moving averages is that it weights all previous periods equally. For example if we have a large amount of historical data, we might want to forecast based on additional data but intuitively we know that last year’s data, while influential, is not as important as the past few quarters. Exponential smoothing is a technique that allows us to accommodate larger data sets by assigning weights to historical data.