LUNCH & LEARN

DEMAND FORECASTING

Dr. Andre Murphy [email protected] 619.591.9715 DEMAND FORECASTING

 Demands & Forecasts

 Forecasting Models DEFINITIONS

. Demand Forecasting: Is the process of making predictions of the future based on past and present data and analysis of trends for a product or service . Demand Management: Includes demand planning, prioritizing and management DEMANDS AND FORECASTS

Demand Types Forecast Types

. Independent . Qualitative

. Dependent . Quantitative INDEPENDENT DEMAND

. Demand for an end item

. Forecasted

Examples: . Uniforms . Tail hook DEPENDENT DEMAND

. Demand for assemblies, components, or ingredients necessary to make an end item

. Calculated based upon Independent Demand

Examples: . Ferrirum M54, C64 steel . Wheel, tire, nut … DEMANDS AND FORECASTS

Demand Types Forecast Types

 Independent  Qualitative

 Dependent  Quantitative DEFINITIONS

. Qualitative Forecasting: When there is little historical data to rely on, such as when establishing sparing for a completely new and unique item, and intuition or expert judgement is required QUALITATIVE FORECAST

Subjective prediction of demand . Often used in absence of historical data . May use mathematical modeling

May include: • Expert judgment • “Delphi” method (panel of SMEs) QUALITATIVE METHODS

Grass Roots: Deriving future demand by asking end-user or an organization closest to the end-user

Market Research: Attempting to identify end-user patterns, emerging trends, and new products for support

Panel Consensus: Deriving future estimations from the synergy of a panel of experts in subject area

Delphi Method: Similar to Panel Consensus but with anonymity

Historical Analogy: Identify another similar commodity to the one being forecasted. DEFINITIONS

. Quantitative Forecasting: When historical data exists and is helpful in calculating future demand or forecasting based on numbers QUANTITATIVE FORECAST

Predicts demand based upon historical data . Often uses math models

Examples of math models . Moving Average . Exponential Smoothing . … QUANTITATIVE METHODS Time Series: Models that predict future demand based on past history

Causal Relationships: Models that use statistical techniques to establish relationships between demand and various outside factors

Simulation: Models that can incorporate some randomness and non-linear effects DEMAND FORECASTING

 Demands & Forecasts

 Forecasting Models 3 Months of Demand Data

Period 1 2 3 4 5 6 7 8 9 10 Forecast

Demand 10 10 15 20 15 10 15 20 25 10

Highest Demand = 15 Lowest Demand = 10 Average Demand ~ 12 10 Months of Demand Data

Period 1 2 3 4 5 6 7 8 9 10 Forecast

Demand 10 10 15 20 15 10 15 20 25 10

Highest Demand = 25 Lowest Demand = 10 Average Demand = 15 FORECAST MODELS Moving Average Forecast: Predicts future need based upon a “rolling” average of historical data

Weighted Moving Average Forecast: Predicts future need based upon weighted historical data . Applies factor to historical period to weight future anticipated demand FORECASTING MODELS

3-Week Moving Average formula Ft+1 = (Dt-2 + Dt-1 + Dt-0)/3 Moving Average Forecast Model

3-Moving Average forecasts based on previous 3 months of data

Period 1 2 3 4 5 6 7 8 9 10 Quarterly 12 15 17 15 13 15 20 Moving Average Historical 10 10 15 20 15 10 15 20 25 10 Demand

4-Month Moving Average forecasts based on previous 4 months of data

Period 1 2 3 4 5 6 7 8 9 10 4-Month Moving 14 15 15 15 15 18 Average Historical 10 10 15 20 15 10 15 20 25 10 Demand Moving Average Forecast Model

6-Month Moving Average forecasts based on previous 6 months of data

Period 1 2 3 4 5 6 7 8 9 10 6-Month Moving 13 14 16 18 Average Forecast Historical 10 10 15 20 15 10 15 20 25 10 Demand

4-Month Moving Average forecasts based on previous 4 months of data

Period 1 2 3 4 5 6 7 8 9 10 4-Month Moving 14 15 15 15 15 18 Average Historical 10 10 15 20 15 10 15 20 25 10 Demand Moving Average Forecast Model

Using a large number of Using a small number of previous periods previous periods Forecast will be slow to Very little smoothing will respond to changing occur conditions Growth and seasonality could Nervous forecasts present result in shortages of supply problems for production and material planners Exponential Smoothing Forecast Model Simple exponential smoothing uses: 1. Use historical data to create initial forecast

2. Use weighted prior Forecast (Ft) and Demand (Dt) to create future forecasts

Future Forecast = Ft+1 = αDt + (1 – α)Ft Where alpha is the exponential smoothing constant Exponential Smoothing Forecast Model

Forecast for future 8 months using exponential smoothing and alpha factor (α) = 0.3

Period Previous Current 1 2 3 4 5 6 7 8

Forecast 15 18 13 14 18 16 12 14 18 22

Actual Demand 10 10 15 20 15 10 15 20 25 10

Forecast for future 8 months using exponential smoothing and alpha factor (α) = 0.6

Period Previous Current 1 2 3 4 5 6 7 8

Forecast 15 18 16 15 17 16 14 15 16 19

Actual Demand 10 10 15 20 15 10 15 20 25 10 EXPONENTIAL SMOOTHING FORECAST MODEL

High α (close to 1) Low α (close to 0) Weight applied to the previous Weight applied to the previous period’s forecast is large period’s forecast is small Responsive to change Slow to respond to changes

Risk of nervous forecasts Stable forecasts EXPONENTIAL SMOOTHING FORECAST

F28 = αD27 + (1 - α)F27 F28 = 0.1 x 900 + (1 - 0.1) x 1000 F28 = 90 + .9 x 1000 F28 = 90 + 900 F28 = 990 Ft+1 = αDt + (1 – α)Ft

• D27 = 900 (actual demand) • F27 = 1,000 (previous period’s forecast for period 27) • Assume that α = 0.1 DEMAND FORECAST ACCURACY

Factors affecting forecast accuracy include; . What level are you calculating – product, family, business unit, total?

. Are you using ship date, invoice date, receipt date to calculate?

. If comparing against contractor/supplier dates, what are they using? DEMAND FORECAST ACCURACY

Location A Location B Total

Forecast Demand 25 75 100

Actual Demand 75 25 100 Forecast Error (%) 67% 100% 0%

Forecast Accuracy(%) 33% 0% 100%

“The Devil is in the Details” In aggregate, the forecast accuracy is 100%, however, if it isn’t the proper mix or location – your customers may not be happy DEMAND FORECAST ACCURACY

Forecast Accuracy: Goodness of demand model fit . How well forecast model predicted actual demand? . How do you measure good fit? What is bias? . Examples of Forecast Error Measurement • Forecast Error, Percent Error • Mean Error, Mean Percent Error • Mean Absolute Error, • Mean Absolute Percent Error, Mean Squared Error • Mean Absolute Deviation, % Mean Absolute Deviation • … FORECAST ACCURACY (CONT.) Depot A Depot B Total Forecast 25 75 100 Example Provided Actual 75 25 100 Percent Forecast 33% 0% (-100%) 100% Accuracy Depot A Depot B Total Alternative Example: Forecast 25 75 100 • Demand volume Actual 75 25 100 weighted using absolute errors: 0% Absolute Error 50 50 100 • Item Average (no weighting – 33% & - Percent 100%): -33.5% Forecast 33% -100% 0% Accuracy 29 DEMAND FORECAST ACCURACY

Forecast Error Measurements Formula

Forecast Error et = Yt – Ft

Percent Forecast Error PEt = (Yt – Ft)/Yt x 100

Mean Absolute Deviation MADt= | | 1 𝑛𝑛 𝑛𝑛 ∑𝑖𝑖=1 𝑥𝑥𝑖𝑖 − 𝑥𝑥̅ Y = Actual Demand F = Forecast e = Error i = iterations from mean DEMAND FORECAST ACCURACY

Forecast Error Measurements Formula

Mean Error MEt = 1 𝑛𝑛 𝑡𝑡=1 𝑡𝑡 Mean Absolute Error MAEt =𝑛𝑛 ∑ 𝑒𝑒| | 1 𝑛𝑛 Mean Squared Error 𝑛𝑛 ∑𝑡𝑡=1 𝑒𝑒𝑡𝑡 MSEt = 1 𝑛𝑛 2 𝑛𝑛 ∑𝑡𝑡=1 𝑒𝑒𝑡𝑡 Y = Actual Demand F = Forecast e = Error DEMAND FORECASTING

 Demands & Forecasts

 Forecasting Models