AN 300 Ch. 3 Test 2

When you notice a horizontal or "flat" pattern for actual demand without trends or seasonalities, what forecast techniques should you choose?

Naive, SMA, WMA, Exponential Smoothing Note: there are just random fluctuations around the average actual demand. Shows NO positive or negative demand.

Calculating the Forecast Error

=Actual - Forecast

A. What does MAPE tell us? B. Benefits? C. When is this measure especially useful?

A. B. Avoids the "error offset" problem by converting all errors to unsigned values (positive). -It also converts the resulting absolute errors to percentages of the actuals before averaging. C. When comparing forecast performance of a model across different SKUs (Stock Keeping Units). - Because everything is measured as percentages with MAPE, native units (e.g., cases of beer versus gallons of milk) of the different stock items no longer matters and the comparison becomes straightforward.

Mean Absolute Percentage Error (MAPE) A. Array formula B. What does the value tell us?

A. =average(abs(forecast errors)/(actual demand)%) Ctrl+Shift+Enter B. Forecast *Deviates* from Actual Demand

Mean Absolute Error/ Deviation (MAD) A. Array formula B. What does the value tell us?

A. =average(abs(forecast errors)) Ctrl+Shift+Enter B. Forecast *Deviates* from Actual Demand

Average Bias/ Average Forecast Error (AFE) A. Array formula B. What does the value tell us?

A. =average(forecast errors) B. Positive = Forecast *Underestimates* actual demand Negative = Forecast *Overestimates* actual demand

Root MSE (RMSE) A. Array formula B. What does the value tell us?

A. =sqrt(MSE) B. Forecast *Deviates* from Actual Demand

Bias/ Cumulative Bias Forecast Error (CFE) A. Array formula B. What does the value tell us?

A. =sum(forecast errors) B. Positive = Forecast *Underestimates* actual demand Negative = Forecast *Overestimates* actual demand

Mean Squared Error (MSE) A. Array formula B. What does the value tell us?

A. =sumsq(forecast errors)/(count(forecast errors)-1) B.Forecast *Deviates* from Actual Demand

A. What does MSE tell us? B. What is the benefit of this metric?

A. Conveys how much your forecast differs from actuals on average, without regard for under/overestimation (i.e., by converting all errors to unsigned values). B. Avoids the "error offset" weakness that CFE/AFE are susceptible to because all errors become positive when squared.

*Simple Moving Average* A. Calculation B. The larger the "n" value ... C. The smaller the "n" value ...

A. Depends on the number of months you are averaging. AP = Actual Demand Period Forecast Period n+1 = (AP1 + AP2 + ... + APn)/n B. The "smoother" the forecast; C. The more "jagged" the forecast. Note: With smaller "n", forecast mimics actual demand behavior better albeit with a "lag" or delay.

*Weighted Moving Average* A. Formula B. In which order do you use the weights? C. All else being the same, larger weights for more recent data will cause WMA to ... D. With WMA, both relative _______ and ________ play a role in how the forecast behaves. 1. Smaller weights and larger "n" value, results in a forecast that is ________. The smoother forecast also results in ___________ errors.

A. Forecast n+1 = (W1 x Actual 1) + (W2 x Actual 2) + ... + (Wn x Actual n) B. If specified, use that. Otherwise, use recency weighting: The largest weight goes to the most recent period and the smallest weight goes to the oldest period. C. Mimic actual demand more closely. D. weights; periods, n. 1. "smoother"; larger

What does AFE tell us?

A. Similar to Cumulative Bias. B. Is a scaled-down representation of CFE(and, yes, CFE is a scaled-up representation of AFE). C. Same virtues and limitations as CFE.

What does CFE tell us? A. If bias turns out to be zero or small positive/negative,... B. If bias turns out to have large positive/negative value,... C. What is a negative about this model?

A. The model is NOT biased or significantly biased. B. The model either consistently under/over-estimates demand by small amounts or occasionally large positive/negative errors. C. It can hide significant positive/negative errors if they cancel each other out. So, CFE value could be small because of this. DECEIVING!!

*Exponential Smoothing* (continued) A. The Larger the "smoothing constant" alpha, ... B. Smaller alpha values provide ... C. What is a benefit?

A. The more closely the forecast tracks reality, albeit with a lag. B. "Smoother" forecasts (and relatively larger forecast errors). NOTE: In the ES method, prior forecasts "feed" into subsequent forecasts. C. The negative impact of a poor Starting Forecast choice in the early forecasts made using ES will becomes less and less as the ES method is applied further into the future.

*Naive forecast* A. Calculation

A. This month's forecast is the previous month's actual.

*Exponential Smoothing* A. How do you choose the first forecast? B. Formula C. Using Excel D. What is alpha?

A. Will be given to you OR choose one of the other techniques (Naive, SMA, WMA) to get the first forecast value. B. Ft+1 = Ft + α(At - Ft) C. Data Analysis --> Exponential Smoothing --> Input Range: Actual Demand column starting with Title --> Check Labels --> Dampening Factor: (1 - alpha) --> Choose Cells where you want the output to be --> OK D. Smoothing Factor

A. What does RMSE tell us? B. What are its benefits?

A. a scaled-down representation of MSE and is represented in native measurement units (e.g., "cases of beer" or "cartons of milk"). B. Becomes a measure that functions just like MAD with the added benefit that large errors have been magnified (by squaring) before averaging and taking the root. Note: It is the largest of the different metrics that use native measurement units -- i.e., CFE, AFE, and MAD.

Which forecast is MORE accurate?

The one with the *lowest* value.