Chapter 9: Forecasting
Long-range forecast (Asset Acquisition)
- Yearly planning bucket
- 3-10 years planning horizon
- New product planning, facility construction, technology
Medium-range forecast (Asset Utilization)
- Monthly/Quarterly planning bucket
- 3 months to 2 years planning horizon
- Seasonal production, inventory, employment, budgeting
Short-range forecast (Asset Execution)
- Weekly/Monthly planning bucket
- 1-26 week planning horizon
- Job scheduling, worker assignments, inventory stocking
Demand Patterns
- Level or Constant
- Randomness
- Trend
- Seasonality
- Cyclical
Level or Constant
Average value is relatively constant over time.
Randomness
Unpredictable movement from one time period to the next.
Trend
Long-term movement up or down in a time series.
Seasonality
A repeated pattern of spikes or drops associated with certain times of the year.
Cyclical
Downward and upward movement of gross domestic product (GDP) around its long-term growth trend.
Random Component
- Unpredictable movement from one time period to the next
- Due to random variation or unforeseen events
- Varying number of daily shoppers
- Rockets win Southwest Division
- Rainy day
- Natural disaster
- Strikes
- Often measured by variance or standard deviation of time series
Trend Component
- Persistent, up or down movement in a time series
- Due to changes in population, technology, product acceptance, etc.
- Trends usually change over time.
- Increase, decrease, remain constant
Patient Forecast
= a + bx, where
- a = current number of patients
- b = trend per month in patients
- x = number of periods to forecast in the future
What is forecast if a = 80, b = 12, and x = 3?
Forecast = 80 + 12(3) = 116 patients
Seasonality Component
- A repeated yearly pattern of spikes or dips in a time series.
- Due to weather, holidays, promotions, etc.
- Seasonal index is expressed as a portion of the average demand in a time period.
Forecasting Approaches
- Qualitative Methods
- Quantitative Methods
Qualitative Methods
- Subjective
- Used when situation is vague and little data exist
- New products
- New technology
- Involves intuition, experience
- e.g., forecasting sales of a new product
Quantitative Methods
- Objective
- Used when situation is stable and historical data exist
- Existing products
- Current technology
- Involves mathematical techniques
- e.g., forecasting sales of commodity products
Qualitative Forecasting Methods
- Market survey
- Build-up forecast
- Panel consensus forecasting
- Delphi method
- Life cycle analogy method
Quantitative Forecasting Approaches
- Time Series Models
- Causal Models
Time Series Models
demand follows a pattern over time.
- Naïve Approach
- Moving average
- Weighted moving average
- Exponential smoothing
- Trend-adjusted Exponential Smoothing
- Linear Regression
Causal Models
demand is predicted by observing environmental factors such as economic indicators
- Linear Regression
- Multiple Regression
Time Series Forecasting Models
A quantitative model that uses a time series to develop forecasts.
- Underlying demand pattern may change over time
- Balance forecast responsiveness versus stability
- Forecasting is typically automated with managerial override.
Time series
A set of observations arranged in chronological order
Assumption
past is a good predictor of the future.
Last Period
Current demand is next period’s forecast.
Naïve Approach
replicates random variation resulting in a highly unstable forecast.
Forecast objective
is to smooth out randomness to find underlying demand pattern.
Moving Average Model
MA t = Moving average forecast prepared at end of period t for use in period t + 1
- n is the number of periods in the moving average
- Smooths data randomness to illuminate data pattern.
- Responsiveness (small n) versus stability (large n)
- Does not consider trend or seasonality
Weighted Moving Average Method
A form of the moving average model that applies varying weights to past observations.
- Highest weights usually placed on more recent past demand
- Responsiveness determined by weight values.
Exponential Smoothing Model
A form of moving average model in which the forecast is calculated as the weighted average of the current period’s actual value and prior forecast.
Features
- Weighted smoothing model with greater weight on most recent data.
- Requires very little stored data and easy to automate.
The general rules for determining the a value:
- Stability
- The greater the randomness in the time series data, the lower the a value.
- The less randomness in the time series data, the higher the a value.
- Responsiveness
- The greater the instability in the underlying data pattern, the higher the a value.
- The less instability in the underlying demand pattern, the lower the a value.
- Select a to minimize forecast error (MAD, MSE, MAPE).
- Smoothing constant is typically 0.05 ≤ α ≤ 0.3
Adjusted Exponential Smoothing Model
An expanded version of the exponential smoothing model that includes a trend adjustment factor.
Measures of Forecast Accuracy
Measures of Forecast Accuracy are used to assess how well a model is performing or to compare multiple forecast models to one another.
- Forecast error for period t
- Running sum of forecast error
- Tracking Signal
Forecast error for period t
FE i = D i – F i
- Note: Right-hand-side values are listed alphabetically. Positive FE i indicates that demand exceeded the forecast.
Running sum of forecast error
- RSFE indicates tendency to over or under forecast
- RSFE should equal 0 in long term.
Tracking Signal
TS provides a measure of the severity of forecast model bias.
- General rule: If - 4 ≤ TS ≤ 4, then the forecast model is performing normally. Otherwise, alternative models should be considered.