Chapter 15 Forecast

unbiased forecast

A forecast that is correct on average, thus an average forecast error equal to zero.

naïve forecasting method

One period of exceptionally high (or low) demand data will likely make the next period forecast highly incorrect. We see no reason to use this method in practice.

Expert panel forecasting

When McDonald's needs to make a forecast for corporate sales, however, there is so much at stake that the costs of generating the forecast simply matter less. So, for forecasts where there is a lot at stake, automated forecasting is typically augmented by expert panels. On such panels, a group of managers share their subjective opinions and try to reach a consensus about a demand forecast.

Automated forecasting

When weather.com makes a prediction for the temperature in Manhattan tomorrow at 9 a.m., it cannot convene an expert panel of meteorologists. Most forecasts in business need to be made millions of times, so they have to be done cheaply, which typically means without human involvement. How many cheeseburgers will customers order in an hour at a particular McDonald's? How many rental cars will be needed on a particular day at a particular airport? Forecasts of these types are created by computers, typically with no human intervention. You might have heard buzzwords such as machine learning and Big Data; both of these stand for sophisticated versions of regression analysis in which computers find out what variables best help make a good prediction.

Mid-term forecasts

are forecasts that are made from the monthly level to the yearly level. They drive capacity-related decisions (recruiting, acquisition of machinery), but also are used for financial planning. In the flu example, this corresponds to making a forecast for the entire flu season so that the right number of nurses can be recruited or the right number of flu vaccines/medications can be produced.

Short-term forecasts

are used to support decisions that are made for short time periods ranging from the daily level to the monthly level. In extreme cases, forecasts might even be made at the hourly level. These forecasts are used to help decisions related to staffing (restaurants have more servers at lunch than in the afternoon) and short-term pricing. They can also be used to predict waiting times and help with scheduling decisions. In the flu example, this corresponds to making a forecast for tomorrow or the next week so that an appropriate number of nurses can be scheduled.

trend

continuing increase or decrease in a variable that is consistent over a long period of time.

A time series-based forecast( extrapolation)

forecast that is obtained based on nothing but old demand data

Long-term forecasts

more than years

exponential smoothing

puts more weight on the recent demand data. This makes it more responsive to changes in demand. Note that to compute the new forecast, all data needed are the latest demand data and the last forecast. However, in a world of Page 503spreadsheets, we argue that this computational simplicity should not be seen as too big of an advantage.

seasonality

repetitive fluctuation over time.

moving average forecast

take care of statistical noise by averaging it out. One has to choose the length of the window over which the average is taken. All periods in the forecast window are weighted equally when computing the new forecast.

Trend formula

y y^ trend 360.000 5.000 1 377 368.400 6.360 2 402 381.480 9.048