The Naive forecast is an extremely simple approach to forecasting future values from a set of observations – you just use the previous time period’s observation to obtain the forecast for the next period. Using a past exam question I cover the method in this question and measure its accuracy using forecast errors, the Mean Absolute Error (MAE), and the Mean Absolute Percent Error (MAPE).
Questions with 'Sample Information' make up the most challenging problems on the test. Learn how to easily identify - based on the given wording - whether or not you are dealing with sample information. This part of the chapter is very similar to the lessons on Probability in Stats 1 (ADMS 2320).
The data in a series of observations taken over time (a time series) can have a lot of variation. Variation leads to unreliable forecasts. To counter this we can use smoothing methods. This question introduces two of them: Exponential Smoothing and the Weighted Moving Average. MAE and MAPE are used to compare their accuracy – plus I’ll show you a third measure of accuracy that sometimes appears on tests: the Mean Squared Error (MSE).
Good news ... You already know this stuff! Expected-Value-without-Sample-Information (EVwoSI) is just the same as Expected Value (EV) covered in my earlier video. I show you how to solve for EV this time using the decision tree, and how that solution becomes a branch of the tree that will help you later with EV-with-SI.
In the previous video we were simply given the value of the Weighted Moving Average as part of the question (Question #2b). Here’s how that value was calculated from the raw data, and how the related MSE was determined from the forecast errors. The Weighted Moving Average appears on a lot of past tests, so it’s important that you can work it out for yourself.
Now it's time to put our sample information to use (also known as: Market Research, Consultant, Report, etc...). The key to solving for the Expected-Value-with-Sample-Information is to draw the decision tree, so I'll show you some tricks to make sure you get the tree set up right.
Remember – Large amounts of variation in the time series data makes forecasting with accuracy difficult to achieve. Some forms of variation (like seasonal) are not random, but rather follow regular patterns that can be removed from the data to increase the reliability of our forecasts. This video shows you how to create a seasonal forecast model from data with seasonal variations and how to use it to make forecasts for time periods falling in different seasons.
We can finally solve by comparing the expected value WITH sample information to the cost that the consultants want to charge ($75,000). The final step is then to write out the Optimal Decision Strategy. This is a list of the correct decisions that should be made as we move from the beginning of the decision tree to the end.
Good News… You already know this stuff! Linear Trend Projection is no different from the linear regression method you learned in Chapter 16 at the beginning of the course. The only thing to keep in mind is that instead of using x and y in the formulas, we replace them to reflect our focus on time series data: x = t (time period), and y = Ft (the forecast value for that time period).
By comparing the value added by the less-than-perfect market research firm (EVSI) to the value of perfect information (EVPI) that we calculated near the beginning of this video series, we can determine how 'efficient' their input really is.