• A statement about the future
• Reduce risk in decision making
• A critical function of business
• Basis for most other planning decisions
• Production planners need for:
• Capacity planning, Scheduling activities, ordering materials, establishing inventory levels, planning shipments

• Qualitative - judgement and opinion
• Expert opinion
• Delphi technique
• Sales-force polling
• Customer surveys

• Quantitative - based on numbers
• Moving averages and weighted moving averages
• Exponential smoothing and trend effects
• Simple and multiple regression
• Trend analysis

• Assumes causal system
• Forecasts rarely perfect because of randomness
• Forecasts more accurate for
  groups than individual
• Forecast accuracy decreases
  as time horizon increases

• Timely
• Accurate
• Reliable
• Meaningful units
• Written
• Simple to understand and use

Steps in the Forecasting Process

• Goal - smooth out statistical noise that obscures systematic movements
• Detect real variations of demand
• Don’t want to over/under produce
• Averaging smoothes out fluctuations

• Naïve forecast - equals the previous period
• Simple
• Easy to prepare
• No cost
• Not highly accurate but may be acceptable

• Uses a number of the most recent actual data values
• MA = (S A)/n
• Number of periods depends on the sensitivity to change required of the forecast - Responsiveness

• Easy to compute
• Easy to understand
• All values weighted equally
• Should "ancient" data have the same weight as recent data

• Assigns more weight to certain values
• WMA = x₁(A) + x₂(A) + x₃(A)
• Tends to be more reflective of the most recent occurrences
• Weight assignment based on experience

• Forecast error = (Actual - forecast)
• New forecast should account for forecast error
• Next forecast = Previous forecast + a(Actual - Previous forecast)
• F_t = F_t-1 + a(A_t-1 - F_t-1)

Forecast accuracy

• Indication of the extent to which the forecast deviates from the observed value
• Comparison of models
• Required in periodic forecasts to ensure that errors are in reasonable bounds

• Forecast error = actual - forecast
• e_t = A_t - F_t
• Positive error = forecast too low
• Negative error = forecast too high
• Helps in selecting forecasting techniques
• Helps in evaluating the technique

• Summarizes historical error performance
• Average of the absolute errors
• MAD = (S ½ Actual - forecast½)/n
• Weighs all errors evenly

• Average of the squared errors
• MSE = (S(Actual -forecast)²)/n-1
• Tends to weigh large errors more heavily