LEC 5: DEMAND FORECASTING

Quantitative method
1) use mathematical techniques
2) based on historical data
3) less accurate as time horizon increases

Forecast accuracy

🤕Qualitative method
based on intuition or judgmental evaluation

Time series forecasting models
use historical data to make prediction

Cause-and-effect forecasting model
based on independent variable to predict demand

Weighted moving average forecast (=sum of old data * their weight)

Exponential smoothing forecast F (t+1) = Ft + alpha * (At-Ft)

Simple moving average forecast (=average of all previous data)

Linear trend forecast Y=B0 =B1*x

Naive forecast (=previous data)

Simple linear regresion forecast Y= b0 + b1*X

Multiple regression forecast Y= b0 + b1X1 + b2X2+...+bk*Xk

Forecast error (et)= At - Ft

Measure of forecastting accuracy

MAPE = average of all |et/At|

MSE= average of all et^2

MAD= average of all (|et|)

RSFE = sum of all et

Sales forces composite: based on experience of sales team to make prediction about customer needs

Customer survey: 5 steps

Delphi method: ased on the results of several rounds of questionnaires sent to a panel of experts

Jury of executive opinion: a meeting of senior management executive to forecast market

carry out the survey (phone, internet, interviews)

4) collect and analyze data

2) choose the target population

5) make forecasts from the results

1) design a forecasting questionnaire

❌ lack of consideration of casual relationship, may not generate accurate forecasts

✅ inexpensive, easy to use

✅ works well when demand is stable, easy to use and understand

❌ affected by random events, respond to change slowly

✅ respond to change quicker than 2 methods above, the weight is based on experience of forecaster

❌ not good for tracking trend changes in data, may lag data due to the nature of average effect

✅ less data required, higher alpha -> more responsive to changes,

❌ lag in actual data bc only partial adjustments to the most recent forecast error can be made

✅ achieve max forecast for key variables, easy to understand and use

❌ accuracy and reliability depend heavily on historical conditions --> not responsive to changes (entrance of new competitor, covid 19)

Tracking signal = RSFE/MAD