Lecture 5 - Forecasts

Question 1

Why do we forecast?

Answer 1

To take action

Question 2

In terms of “Sales & Operations Planning”, What does it mean and what is the objective?

Answer 2

• WHAT: Process that integrates demand, supply, and financial planning into one business plan.
• OBJECTIVE: Align teams by matching demand and supply in the most desirable product portfolio and mix to maximize sales and profit.

Question 3

What are the 4 dimensions to set up a “Demand Forecasting Process”?

Answer 3

1. Granularity
2. Temporality
3. Process
4. Metrics

Question 4

What is meant by “Granularity”?

Answer 4

How we should forecast is dependent on Material, Localization, Time buckets and Horizon.

Question 5

There are 3 types of hierarchical forecasts, which ones?

Answer 5

1. Top-down
2. Middle-out
3. Bottom-up

Question 6

What is meant by “Bias” and “Accuracy” and what is the trade-off between them?

Answer 6

Bias: tendency to over- or under forecast.

Accuracy: amplitude of the forecast error

Rather have Biased and Accuracte results than unbiased and non-accurate results. The first one enables you to take action.

Question 7

What is the formula for “Error”?

Answer 7

error = Forecast - Demand –> e = f - d

Question 8

What is the formula for “Bias”?

Answer 8

Bias = [1/N] * Sum of e

In %: We take, (sum of e) / (sum of d)

Question 9

What is meant by “MAE” and what’s the formula?

Answer 9

Mean Absolute Error.

MAE = [1/N] * Sum of | e |
MAE in % = (Sum of | e |) / (sum of d)

Question 10

What is meant by “MAPE” and what’s the formula?

Answer 10

Mean Absolute Percentage Error.

MAPE = [1/N] * (Sum of | e |) / (sum of d)

Notice that the difference is that we take the average of MAE.

Question 11

What is meant by “RMSE” and what’s the formula?

Answer 11

Root Mean Squared Error. Penalizes bigger errors.

RMSE = [ [1/N] * (Sum of e^2) ]^(1/2) ]

RMSE in % = [ [1/N] * (Sum of e^2) ]^(1/2) / [ [1/N]*(sum of d) ]

Question 12

Which of the 4 metrics is the best one to use?

Answer 12

All of them are good, it depends on our situation.

Question 13

MAE vs RMSE, which method is best to use to prevent the effect of outliers?

Answer 13

MAE is biased but accurate and then RMSE is more sensitive to outliers. So, MAE.

Question 14

How do we know if our model is doing a great job?

Answer 14

We compare it to benchmarks.

Question 15

What is meant by “Efficacy” and “Efficiency”?

Answer 15

for instance:
Efficacy: You do not want any team member to spend time editing the forecast while not improving it.

Efficiency: You do not want any team member to spend too much time editing the forecast.

Question 16

What should we do to forecast demand by excellence?

Answer 16

We should achieve efficacy, efficiency, and focus. The forecasting KPI should be done on the weighted MAE + weighted Bias.

Question 17

When should we use “Judgemental Forecasting”?

Answer 17

When there is present information about actions in the future that the models can’t forecast.

Examples:

• I know my main client expects lower sales than usual. I will reduce the forecast.
• I will input demand for this new product.
• We will increase pricing next month. I should reduce the forecast.

Question 18

What are 2 sources of “judgemental Intentional bias”?

Answer 18

1. Misalignment of Incentives

2. Power and Influence

Question 19

What is the “Cobra Effect”?

Answer 19

Cobra effect – incentivizing people with cash could also go wrong.

Question 20

What are the sources of “judgemental Unintentional bias” and what is meant by them?

Answer 20

• Cognitive Bias - (unconsciously) we are all looking for information supporting our current beliefs and ideas. While avoiding contradictory information.
• Anchoring Bias - we think by comparison. The first number we think of will influence our thought process (even if it is irrelevant).
• Apophenia - we are hard-wired to see patterns (when there is no specific link)

Question 21

What is meant by “Time series forecasting”?

Answer 21

The forecast is an extension of the historical demand.

Question 22

Give a brief explanation of each of the theoretical components of demand - there are 5 of them

Answer 22

The five theoretical components of demand are the trend, the cyclical, the seasonal, the average and the random components.

1) The average component of demand is the level around which demand varies over a period of time.
2) The trend is the persistent (long term) pattern of demand. It may depend on many factors, including the Product Life Cycle phase that the product is in.
3) The cyclical component of demand is a multiple year pattern that arises from economic, political or other factors.
4) The seasonal component of demand is the shorter-term (one year or less) pattern of demand.
5) The random component of demand is the residual part of demand. It is the erratic difference between the four previous components and actual demand (the noise).

Question 23

What is the formula for Exponential Smoothing with a Trend?

Answer 23

1) F_t = a*A_(t-1) + (1-a) [ (F_(t-1)+T_(t-1) ]
2) T_t = B(F_t - F_(t-1)) + (1-B) * T_(t-1)
3) FIT_t = F_t + T_t

We start with 1) and 2) the in 3) is where all work flows out into the forecast that we will use. Then repeat.

Question 24

What are the respective properties of CFE, MAE, MSE and MAPE?

Answer 24

The CFE does not reward precision. It only rewards the absence of bias.
The MAE rewards median forecasts.
The MSE rewards average forecasts.
The MAPE has a tendency to undershoot demand by overweighing periods with low demand.