An estimate of the future level of some variable, used to predict future event, and determine
- Long-term Capacity Needs
- Yearly Business Plans
- Short-term OSC Activities
Forecasting
3 Common Forecast Types
- Demand
Overall market / Firm-level - Supply
# of current suppliers/supply aggregated/tech & political trends - Price
$ of key materials & srv
Underlying basis of all business decisions
- Production
- Inventory
- Personnel
- Facilities
Forecasting Time Horizons
- Short-range
3 mo. to 1 yr
more accurate
purchasing/job schedule/workforce lvl/job assgmt/production lvl - Medium-range
3 mo. to 3 yrs
sales & production planning/budgeting - Long-range
3+ yrs
new product planning/facility location/r & d
In product life cycle (Intro/Growth/Maturity/Decline), which require longer forcasts?
Intro & Growth
As product passes thru life cycle, forecasts are useful in projecting
- Staffing levels
- Inventory levels
- Factory Capacity
SIF
4 Laws of Forecasting
Law1: Almost always Wrong
Law2: Near term Accurate↑
Law3: Groups of G/S Accurate↑
all cars vs green cars
Law4: No Substitute for calculated values
Features of Forecasts
- Causal system assumed
- Randomness makes forecast imperfect
- Accurate for Groups
- Time Horizon↑Accuracy↓
Select Forecasting Method if
based on intuition/informed opinion & little data
Qualitative techniques
Select Forecasting Method if
use measurable/historical data
Quantitative techniques
5 Qualitative techniques
- Market Surveys
questionnaires solicit/time+$ ↑ - Build-up
expert familiar w/ mkt segment - Life Cycle Analogy
for new G/S - Panel Consensus
experts work together - Delphi
experts work individually → modify w/ shared info
MLBPD
2 Quantitative models
& subs
- 時序Time series models
a. Last Period
b. Moving Avg
c. Weighted Moving Avg
d. Exponential Smoothing
e. Linear Regression - 因果Causal models
a. Linear Regression
b. Multiple Regression
A class of
Quantitative Forecasting
modeled as function
other than time
Causal Forecasting models
e.g. Variable/Cause
Drought relief/Rainfall
Refinancing/Interest rates
Food eaten/# of guests
3 Major Questions for Life Cycle Analogy method
- How Long for each stage?
- How Rapid Growth/Decline be?
- How Large is Overall/Maturity ph. Demand?
Demand Patterns of
Time Series Forecast
- Randomness
Unpredictable movement - Trend
long-term movement - Seasonality
repeated spike/drop
Current Demand
= Forecast Next Period
Time Series/Quantitative
Last Period model
Average of a Set of Recent Values generates smoothed forecast
Time Series/Quantitative
Moving Average model
(or Smooting model)
x-Period : avg of prev x period data
random↑⇒smooth/delay↑: for x↑
Forecast @next period =
Weighted Avg Value & Forecast @current period
Exponential Smoothing model
F₂=αD₁ + (1-α)F₁
0≤α ≤1 (smoothing constant)
random ↑ ⇒ α ↓
Adjusted Exponential Smoothing model
AF₂ = F₂+T₂
AF₂ : Adj. Fc @next period
F₂ : UnAdj. Fc @next period
= αD₁+ (1-α)F₁
T₂ : Trend Factor @next period
= β(F₂-F₁) + (1-β)T₁
T₁ : Trend Factor @curr. period
β : Smoothing Constant for Trend Factor
Forecast Variables expressed as Linear Function or independent variable
Causal/Quantitative
Linear Regression
y=a+bx
a: intercept term=
b: slope coefficient (trend)
x: independent variable
(x₀ = mean x)
y: forecast (dependent)
(y₀ = mean y)
∑xy - [ (∑x∑y)/n ] b=----------------------
∑x² - [ (∑x)² / n ]4 Steps to develop
Seasonality Adjustments
- Unadjusted forecast model
- Ratio of “Demand÷Forecast”
∙ if Ratio<1 ⇒ OverForecasted
∙ if Ratio>1 ⇒ UnderForecasted - Seasonal Index (si)
= Ratio (in Mo.s or Qtrs) - Adjusted Forecast model
= si × Unadjusted
FE
Forecast Error
Fc Accuracy Measure
= Demand - Forecast
= D-F
MFE
Mean Forecast Error
Fc Accuracy Measure
= ∑FE / n
= ∑(D-F)/n
Measures bias/propensity of model to under-/over- forecast
(unbiased if MFE=0, even having large errors)
AD
Absolute Deviation
Fc Accuracy Measure
= |FE|
= |D-F |
