Supply Chain: Demand Forecasting

• IE 013 Supply Chain Management presentation is about Demand Forecasting
• Demand forecasting is predicting the future based upon specific assumptions

Intended Learning Outcomes

• Define demand forecasting
• Explain its role in the supply chain
• Identify and explain qualitative and quantitative forecasting techniques
• Define and assess forecast accuracy
• Solve real-life business problems related to demand forecasting and accuracy

Topics Covered

• Introduction to Demand Forecasting
• Time Series Forecasting Models
• Cause & Effect Forecasting Models
• Forecast Accuracy
• Forecasting Software and Websites

What is a Forecast?

• A prediction/projection of the future, based on specific assumptions.

Demand Forecasting

• An important element of demand management providing an estimate of future demand
• Acts as a basis for planning and sound business decisions
• It's both an art and a science
• There will always be deviation between predicted and actual demand
• The goal of forecasting is minimizing deviation

Types of Forecasting Techniques

Qualitative

• Based on intuition or judgmental evaluation
• Generally used when data is limited, unavailable, or not currently relevant

Qualitative Methods

• Jury of executive opinion
• Delphi method
• Sales force composite
• Customer surveys

Quantitative

• Uses mathematical techniques based on historical data
• Can include causal variables to forecast demand

Quantitative Methods

• Time series forecasting models
• Naive forecast
• Simple moving average
• Weighted moving average
• Exponential smoothing
• Linear trend
• Cause and effect models
• Simple linear regression
• Multiple regression

Qualitative Forecasting Methods

Jury of executive opinion

• A group of senior management executives with knowledge about the market, competitors, and the business environment collectively develops the forecast

Salesforce composite

• Forecast is generated based on the sales force's market knowledge and customer needs estimates
• Reliable due to sales force proximity to consumers, but individual biases can impact effectiveness

Delphi method

• Internal and external experts are surveyed in several rounds about future events and long-term demand forecasts, without meeting physically
• Answers accumulate after each round, are summarized, and sent back to experts, who can modify responses based on group's summary. Iterations continue until consensus

Consumer survey

• A questionnaire is designed to get consumer inputs on buying habits, new product ideas and opinions on existing products
• Administered via telephone, mail, internet, or personal interviews
• Data is analyzed using statistical tools and judgement to derive results

Quantitative Forecasting Methods

Time series forecasting

• Assumes the future is an extension of the past, using historical data to predict future demand

Cause & effect forecasting

• Assumes one or more factors (independent variables) relate to demand so that they may be used to predict future demand

Components of a Time Series

Trend variations

• Increasing or decreasing movements over many years
• Influenced by population growth, population shifts, cultural changes, and income shifts

Cyclical variations

• Wavelike movements longer than a year
• Influenced by macroeconomic and political factors like the business cycle

Seasonal variations

• Peaks and valleys repeating over a consistent interval like hours, days, weeks, months, years, or seasons
• Affecting the performance of seasonal companies

Random variations

• Unexpected or unpredictable events such as natural disasters (hurricanes, tornados, typhoons, fire), strikes, and wars

Time Series Forecasting Models

Naive forecast

• Predicts the next period using actual demand from the immediate past period
• Formula: Ft+1 = At
• Ft+1: forecast for the period t+1
• At: actual demand for period t

Simple moving average forecast

• Uses historical data and works well when demand is fairly stable over time
• When n = 1, simple moving average is the naïve forecast
• Formula: Ft+1 = (At + At-1 + At-2 + ... + At-n) / n
• Ft+1: forecast for the period t+1
• At: actual demand for period t
• n: number of periods to calculate the moving average

Weighted moving average forecast

• Uses unequal weights for each data point, giving more emphasis to recent data to reflect current demand pattern changes
• Weights assigned should be non-negative and total to 1
• Formula: Ft+1 = (At Wt) + (At-1 Wt-1) + (At-2 Wt-2) + ... + (At-n Wt-n)
• Ft+1 = forecast for the period t+1
• At = actual demand in period t
• w t= weight of period t
• n = number of periods used to calculate the moving average

Exponential smoothing

• Sophisticated weighted moving average technique
• Forecast for the next period's demand is the current period's forecast adjusted by a fraction of the difference between the current period's actual demand and forecast
• Requires less data than the weighted moving average since it only needs two data points
• Simple and most suitable to use for data showing little trend or seasonal pattern
• Formula: Ft+1 = Ft + α(At - Ft)
• Ft+1: forecast for the period t+1
• Ft: forecast for the period t
• At: actual demand for the period t
• α: smoothing constant (0 <= α <= 1)

Linear trend forecast

• Can be estimated by using simple linear regression to fit a line to data occurring over time
• Also referred to as simple trend model
• Trend line is determined using the least squares method, which minimizes the sum of squared deviations to determine the linear equation characteristics
• The linear trend equation is: ŷ = bo + b₁x
• Ŷ: forecast or dependent variable
• x: time variable
• bo: intercept of the vertical axis
• b₁: slope of the trend line
• The coefficients b & b₁ are calculated as:
• b₁ = (nΣ(xy) - Σx Σy) / (n Σx² – (Σx)²)
• b₀ = (Σy – b₁ Σχ) / n
• x = independent variable values
• y = dependent variable values
• n = number of observations

Cause and Effect Models

Simple linear regression forecast

• Used with one explanatory variable
• Similar to the linear trend forecast except the "x" variable is an explanatory variable of demand and not just a time period
• The regression equation is calculated as: Y = bo + b₁x
• Y: forecast or dependent variable
• x: explanatory or independent variable
• bo: intercept of the vertical axis
• b₁: slope of the trend line
• The coefficients b & b₁ are calculated as follows:
• b₁ = (nΣ(xy) - Σx Σy) / (n Σx² – (Σx)²)
• b₀ = (Σy – b₁ Σχ) / n
• x: independent variable values
• y: dependent variable values
• n: number of observations

Multiple regression forecast

• Used when using several explanatory variables to predict the dependent variable
• This works well when the relationship between demand (dependent variable) and several other factors impacting demand are strong and stable over time
• Determining the parameters of a multiple regression equation is complex
• Can be done with Excel, SAS, or SPSS
• The multiple regression equation is defined as: Ŷ = bo + b1x1 + b2x2 + ... + bkxk
• Ŷ: forecast or dependent variable
• xk: kth explanatory or independent variable
• bo: constant
• b₁: regression coefficient of independent variable xk
• Results must be checked for goodness of fit and statistical significance before estimating

Forecast Accuracy

• The ultimate goal of any forecasting activity is to have an accurate and unbiased forecast
• Costs related to forecast errors can include lost sales, safety stock, unsatisfied customers, and loss of goodwill
• Forecast accuracy can determine forecast bias within acceptable control limits
• Forecast accuracy can compare two sets sets of forecasts

What is a Forecast Error?

• The difference between the actual quantity and the forecast:
• Et = At - Ft
• e t= forecast error
• At = actual demand for period t
• Ft = forecast for period t

Measures of Forecast Accuracy

MAD - Mean Absolute Deviation

• An indicator of forecast accuracy based on an average of the absolute value of the forecast errors over a given period of time
• Indicates, on the average, how many units the forecast is off from the actual data
• Formula: MAD = (|e1| + |e2| + ...+ |en|) /n
• e = forecast error
• n = number of periods

MAPE - Mean Absolute Percentage Error

• An indicator of forecast accuracy based on true magnitude of the forecast error
• The average of the monthly absolute forecast error multiplied by 100
• Indicates, on average, what % the forecast is off from the actual data
• Formula: MAPE = [(|e1/A1|+ |e2/A2|+...+ |en/An|)/n] X 100
• e: forecast error
• A: actual demand
• n: number of periods

MSE - Mean Square Error

• An indicator of forecast accuracy, and is the average of the squares of the forecast errors
• Penalizes large errors more than small errors
• Formula: MSE = (e₁² + e₂² + en²) / n
• e = forecast error
• n = number of periods

RSFE - Running Sum of Forecast Errors

• An indicator of bias in the forecast
• Measures how a forecast tends to be consistently higher or lower than actual demand
• A positive RSFE means the forecasts are less than actual demand
• A negative RSFE means the forecasts are greater than actual demand
• Formula: RSFE = e₁ + e₂ + e₃ + ..+ en
  • e: forecast error
  • n: number of periods

Tracking Signal

• Used to determine if forecast bias falls within acceptable limits
• Acceptable control limits for forecast bias is within + or - 3 (+/-3)
• Tracking signal = RSFE / MAD

Useful Forecasting Websites

• Institute of Business Forecasting and Planning https://ibf.org/
• International Institute of Forecasters https://forecasters.org/
• Forecasting Principles http://www.forecastingprinciples.com/
• Stata http://www.stata.com/

Some Forecasting Software

• Business Forecast Systems, Inc. http://www.forecastpro.com/
• John Galt http://johngalt.com/
• Mi9 Retail https://mi9retail.com/
• Avercast, LLC. http://www.avercast.com/
• SAS https://www.sas.com/

Cloud-based Forecasting

• Using supplier hosted or software-as-a-service (SaaS) advanced forecasting applications that are provided to companies on subscription basis
• Accomplished with state-of-the-art time series forecasting algorithms using seasonal and cyclical adjusting models or artificial intelligence-based expert systems