Forecasting

What’s Forecasted in the Supply Chain?

  • Demand, sales, or requirements,

  • Purchase prices.

  • Replenishment and delivery times.

Nature of Forecasting

  • Spatial vs. Temporal Demand.

  • Lumpy vs. Regular Demand.

  • Dependent (Derived) vs. Independent Demand.

Forecasting Horizons

  • Long-Term

    • Greater than 3 years.

    • Used for Strategic Planning - Production capacity, Inventory levels, Product Range, etc.

  • Midrange

    • One to three years.

    • Used for budgeting and sales planning.

  • Short-Term

    • In terms of months.

    • Used for Tactical planning - Production Schedules, Logistics Plans, Material Planning etc.

Forecasting Steps

  • Identify what to forecast

    • Level of detail, units, of analysis & time horizon required.

  • Identify what data is available

    • Identify needed data & whether it’s available.

  • Select and test a forecasting model

    • Cost, ease of use & accuracy.

  • Generate the forecast

  • Monitor forecast accuracy over time

Identifying Demand Patterns

Typical Time Series Patters: Random

Typical Time Series Patters: Random with Trend

Typical Time Series Patterns: Random with Trend and Seasonal

Typical Time Series Patterns: Lumpy

Forecasting in High Certain Environments

Managing Highly Uncertain Demand

  • Delay forecasting as long as possible.

  • Seek information directly from customers.

  • Prioritize supply by product’s degree of uncertainty (supply to the more certain products first).

  • Apply the principle of postponement to the most uncertain products (delay committing to a final product form until an order is received).

  • Create flexible supply to changing demand (alter capacity and output rates through subcontracting, computer technology, multi-purpose processes, etc.)

  • Delay supply response until demand becomes clear.

  • Be able to respond quickly to uncertain demand levels.

Managing Highly Uncertain Demand: Collaborative Forecasting

  • Demand is lumpy or highly uncertain.

  • Involves multiple participants each with a unique perspective—“two heads are better than one.”

  • Goal is to reduce forecast error.

  • The forecasting process is inherently unstable.

Determine the Best Forecasting Method

Qualitative Methods [Used when situation is vague & little data exist]

  • Market Research

    • Uses customer surveys and interviews to determine customer preferences.

  • Panel Consensus

    • Executives meet and develop a forecast together.

  • Delphi Method

    • Develop a consensus forecast among a group of experts.

Quantitative Methods [Used when situation is ‘stable’ & historical data exist]

  • Regression Analysis

  • Time-Series Analysis

    • A time series is a sequence of observations of a process over points in time, e.g., monthly sales, annual enrollment, weekly production, etc.

    • It is used when historical data contains patterns that can be exploited.

    • Average demand for the period, trend, seasonal elements, cyclical elements, random variation.

Time Series Forecasting Methods

Demand pattern has neither trend, nor seasonal or cyclical effects

  • Simple (or weighted) moving average.

  • Exponential smoothing.

Demand pattern has only trend

  • Exponential Smoothing with Trend.

  • Linear Regression (Forecast given the parameters).

Demand pattern has seasonal effects

  • Exponential smoothing with trend and seasonality.

  • Decomposition of a time-series (Forecast given the parameters).

Moving Average

  • Used when data has no trend, cycle or seasonal characteristics.

  • Used often for smoothing.

    • Provides overall impression of data over time.

  • Notation

    • Ft=Forecast in period t

    • At=Actual data in period t

  • Simple Moving Average

Moving Average (MA) Example

Weighted Moving Average (WMA)

Similar to Simple MA except different weights to actual demand.

Weights based on intuition, expert judgement, experience.

  • Values between 0 & 1 that sum to 1.0

Exponential Smoothing

  • Form of WMA

  • Only 3 pieces of information needed for forecasts unlike WMA.

    • Previous period’s forecast: Ft-1

    • Previous period’s actual demand: At-1

    • Smoothing constant or response rate: α

      • α is between 0.0 and 1.0

      • Higher the value, more the response

  • Forecast

    • - Ft=Ft-1+α (At-1-Ft-1)

Forecasting Error Metrics

Range of the Forecast

If forecast errors are normally distributed and the forecast is at the mean of the distribution, i.e.,

A forecast confidence band can be computed. The error distribution for the level-only model results is:

Selecting a Forecast Model

  • Need to look at cost and accuracy.

    • Cost of Data: Software, Personnel, Amount of Data, etc..

    • Accuracy: MAD, MAPE, etc.

    • Trade-off between cost and accuracy - Why?

  • Comparing models based on accuracy

    • Which is the better model?

Forecasting Exercise

Forecasting Summary

  • Forecasts are uncertain, imprecise, and unavoidable.

  • Long-term forecasts are less accurate than short-term forecasts (forecast horizon is important).

  • Aggregate forecasts are more accurate than disaggregate forecasts.

  • Forecasts are made for many purposes, using many methods.

  • Selecting what is forecast, what model to use must be done carefully.