DEMAND FORECASTING
📘 1. Introduction
🔹 What is Forecasting?
Forecasting is the process of predicting future events based on historical data. It forms the foundation
of planning in various functions like production, finance, HR, marketing, and operations.
🔹 Why Is Forecasting Important?
● It supports better resource allocation.
● Forecasts are used to plan demand, inventory, staffing, and capacity.
● Even though forecasts are not always accurate, good forecasts reduce uncertainty and support
decision-making.
🔹 Types of Forecasts by Time Horizon:
Time Horizon Duration Use Cases
Short-range < 3 months Job scheduling, daily operations
Medium-range 3 months–2 years Sales and production planning
Long-range > 2 years Product development, R&D
📘 2. Qualitative and Quantitative Forecasting
🔹 A. Qualitative Forecasting Methods (Subjective Judgment)
1. Executive Judgment
○ Decisions made based on experience and insights of top managers.
○ Fast, simple, but can be biased.
2. Sales Force Composite
○ Salespersons estimate future sales in their regions.
○ Forecasts are compiled and reviewed for consistency.
3. Market Research/Surveys
○ Uses customer surveys, focus groups, and panels.
○ Effective for new product forecasting.
4. Delphi Method
○ Panel of experts participate anonymously.
○ Repeated rounds of questions and feedback until consensus is achieved.
○ Avoids groupthink and dominant voices.
🔹 B. Quantitative Forecasting Methods (Data-Based)
Time Series Models: Use historical data assuming past patterns will continue. components of a time series:
● Trend: Long-term upward/downward movement.
● Seasonality: Repeating short-term patterns.
● Random Variation: Unpredictable factors.
➤ 1. Naive Method
● Forecast = Last period's actual value
● Example: If May sales = 48, then June forecast = 48.
● Best for data with no trend or seasonality.
➤ 2. Moving Average
a. Simple Moving Average (SMA)
● Average of the last n periods:
● Use: When demand is stable and has no trend.
b. Weighted Moving Average (WMA)
● Assigns more weight to recent data:
○ Weights must sum to 1.
○ Example: Weights = 3/6, 2/6, 1/6 for the last three months.
➤ 3. Exponential Smoothing
● More responsive to recent data.
● Formula: OR
● α = Smoothing constant (0 < α < 1).
○ Higher α = More responsive to recent changes.
● Requires only one prior forecast and actual data point.
➤ 4. Regression Models
Used when a dependent variable (e.g., demand) is influenced by independent variables (e.g., price,
advertising).
● Simple Linear Regression Equation: y= a + b xy
Where:
○ y = dependent variable (e.g., demand)
○ x = independent variable
○ a = intercept
○ b = slope
● Example: Predict tire sales based on car sales
📘 3. Measures of Forecasting Errors
Forecast accuracy is evaluated using error metrics:
Metric Formula Description
MAD (Mean Absolute At= Actual value at time period t
Deviation) Ft= Forecast value at time period t
| At − Ft | = Absolute error
n = Number of time periods
MSE (Mean Squared Emphasizes larger errors due to
Error) squaring.
RMSE (Root Mean Same as MSE, but in original units.
Squared Error)
Tracking Signal (TS) Indicates bias in the forecast.
🔹 Good Forecast Characteristics:
● Low MAD, MSE, RMSE.
● Minimal bias (low tracking signal).
● Reflects understanding of the system.
● Adaptable and tested for accuracy.
✅ Summary Chart: Forecasting Methods
Method Type Best For Pros Cons
Naive Quantitative No trend Simple Not accurate
SMA Quantitative Stable demand Smooths data Lags trend
WMA Quantitative Recent changes Weighs recent values need good weight selection
Exponential Quantitative Trend / randomness Easy to update Choosing α is tricky
Smoothing
Regression Quantitative Causal relationships Models impact of factors Needs accurate data
Delphi, Market Qualitative New products, Human insight Subjective, time-consuming
Research uncertain markets
In ppt but not in syllabus:
📘 1. Decomposition of Time Series (Extended Topic)
While your PPT shows a time series graph with trend, seasonality, and random variation, it doesn’t
explicitly define or decompose them.
🔹 Time Series Components:
1. Trend (T): Long-term increase or decrease in the data.
2. Seasonality (S): Regular periodic fluctuations (e.g., sales rising in December).
3. Cyclic (C): Long-term wavelike patterns (often economic cycles).
4. Irregular/Random (I): Unpredictable residual variation.
🔹 Additive vs. Multiplicative Models:
● Additive Model:
● Multiplicative Model:
📘 2. Seasonality Index Calculation (Missing)
If your data shows seasonal patterns (e.g., quarterly or monthly sales), you may need to calculate
seasonality indices.
🔹 Steps to Calculate Seasonal Index:
1. Find the average for each season (e.g., average Q1 sales across years).
2. Calculate the overall average of all data points.
3. Divide each seasonal average by the overall average.
Seasonal Index = Seasonal Average / Overall Average
Use these indices to adjust forecasts for seasonality.
📘 3. Linear Regression – More Details
While the PPT gives the formula: y = a + bxy
Here’s how to calculate a and b:
🔹 Formulas:
Where:
● x is the independent variable (e.g., advertising spend)
● y is the dependent variable (e.g., demand)
● n is the number of data points
📘 4. Multiple Regression (Optional Advanced)
Used when more than one independent variable affects the forecast. Example: Forecasting sales using
advertising, price, and income level.
📘 5. Causal Models (Related to Regression)
Causal models assume a cause-effect relationship between variables.
● Examples:
○ Ice cream sales depend on temperature.
○ Tire sales depend on car sales.
Regression is a type of causal model. More advanced ones include:
● Econometric Models
● Input-output Models
✅ Summary of Additional Concepts (Handy Table)
Concept Description Use Case
Time Series Decomposition Separates data into trend, seasonality, For analyzing
cycle, and noise historical behavior
Seasonal Index Quantifies seasonal effect To adjust forecasts for
seasonality
Regression Calculation Formula for a and b For creating linear
models
Multiple Regression More than one factor affects y Advanced forecasting
Causal Models Uses external variable impact Strategic forecasting
