Study guide
6.1 Demand Forecasting Methods: Moving Averages, Exponential Smoothing & Error Metrics
PE Industrial and Systems Exam
Track your progress as you read
Finished sections get a check mark.
1 Exam Overview, Strategy & Systems Foundations
1.1 PE Industrial Exam Format, Official Specification & Reference Handbook Strategy 1.2 System Analysis Tools, Process Definition, Requirements Engineering & QFD 1.3 Performance Measures: Output, Productivity, OEE, Cost & Environmental Sustainability 1.4 Process Types (Discrete, Continuous, Manufacturing, Service) & Value Analysis/Engineering
2 Optimization, Queueing, Simulation, Networks & Markov Chains
2.1 Linear Programming Formulation, Graphical Solutions & Sensitivity Analysis 2.2 Queueing Theory Models (M/M/1, M/M/s) & Little's Law Applications 2.3 Discrete-Event Simulation, Random Variates & Model Validation 2.4 Graph Theory & Network Models: Shortest Path, Minimal Spanning Tree & Maximum Flow 2.5 Markov Chains: Transition Matrices, Steady-State Probabilities & Absorbing States 2.6 Decision Trees, Expected Value & Bottleneck Analysis (Theory of Constraints)
3 Project Management & Planning
3.1 Project Scheduling with CPM and PERT: Critical Path, Float, Probability & Crashing 3.2 Project Planning & Control: Scope and WBS, Gantt Charts, Critical Chain (CCPM), Risk & Earned Value
4 Facility Layout, Space, Capacity & Material Handling
4.1 Systematic Layout Planning (SLP) & Activity Relationship (REL) Analysis 4.2 Workstation Design, Aisle Sizing & Space Requirements Planning 4.3 Capacity Analysis: Machine Requirements, Scrap Allowances & Personnel Requirements 4.4 Material Handling Equipment Selection, Unit Loads & Handling Principles
5 Site Location, Warehousing & Facility Economics
5.1 Facility Location Models: Center of Gravity & Minisum / Minimax Distance 5.2 Warehousing Systems, Storage Assignment & Order Picking Operations 5.3 Facility Life-Cycle Economics, Replacement Analysis & Rearrangement Costs
6 Demand Forecasting, Production Planning, MRP & JIT
6.1 Demand Forecasting Methods: Moving Averages, Exponential Smoothing & Error Metrics 6.2 Aggregate Planning Strategies & Master Production Scheduling (MPS) 6.3 Material Requirements Planning (MRP), Bill of Materials & Capacity Planning 6.4 Just-in-Time & Lean Production Planning: Pull Systems, Kanban Sizing, Heijunka & SMED
7 Inventory Systems, Scheduling, Logistics & Transportation Modes
7.1 Deterministic Inventory Control: Classical EOQ, EPQ & Quantity Discounts 7.2 Stochastic Inventory Management: Safety Stock, Continuous / Periodic Review & Newsvendor 7.3 Production Sequencing, Dispatching Rules & Johnson's Algorithm 7.4 Supply Chain Logistics, Transportation Simplex & Distribution Network Design 7.5 Transportation Modes, Freight Economics & the Transshipment Model
8 Engineering Economics, Break-Even Analysis & Costing Systems
8.1 Equivalence, Cash Flow Analysis & Discounted Cash Flow Metrics (PW, FW, EAC) 8.2 Rate of Return Analysis (IRR, Incremental IRR) & Benefit-Cost (B/C) Ratios 8.3 Asset Depreciation (MACRS, Straight-Line), Tax Effects & Replacement Decisions 8.4 Break-Even Analysis, Technical Capability Assessment & Costing Systems (Traditional, ABC & Standard Costs)
9 Methods Engineering, Work Measurement, Learning Curves, Line Balancing & Process Planning
9.1 Methods Analysis, Process Charting & Motion Study (Therbligs, 5S) 9.2 Worker–Machine (Man–Machine) Charts, Multiple Activity Analysis & Synchronous Servicing 9.3 Time Study Fundamentals, Performance Rating & Allowance Factors 9.4 Predetermined Motion Time Systems (PMTS, MTM, MOST) & Work Sampling 9.5 Learning Curves: Unit and Cumulative-Average Models, Learning Rates & Applications 9.6 Assembly Line Balancing, Precedence Constraints & Efficiency Heuristics 9.7 Manufacturing & Service Process Planning: Operation Selection, Sequencing, Route Sheets, Fixturing & Mistake Proofing
10 Ergonomics, Human Capacity, Occupational Safety & DART Rates
10.1 Anthropometry, Percentile Design, Workstation Ergonomics & Link Analysis 10.2 Industrial Biomechanics, the NIOSH Lifting Equation & Lifting Aids 10.3 Limits of Human Capacity & Risk-Factor Tools: Work Physiology, Strength, Range of Motion, Vision, RULA & Slip Resistance 10.4 Environmental Ergonomics (Noise, Heat, Illumination), Human-Machine Systems & OSHA Safety 10.5 Safety Codes, Standards & Voluntary Guidelines (OSHA, ANSI, NIOSH, MIL-STD) and DART Rate Calculations
11 Statistical Process Control & Process Capability
11.1 Statistical Process Control for Variables (X-bar/R and X-bar/s Charts) 11.2 Statistical Process Control for Attributes (p, np, c, and u Charts) 11.3 Process Capability Indices (Cp, Cpk, Cpm, Pp, Ppk) & Sigma Levels
12 Design of Experiments, Acceptance Sampling & Quality Systems
12.1 Design of Experiments: 2^k Factorials, ANOVA & Taguchi Robust Design 12.2 Acceptance Sampling: OC Curves, AQL, LTPD, AOQL, Z1.4/MIL-STD-105E & Dodge-Romig 12.3 Quality Systems (Deming, TQM, ISO 9000, Benchmarking), Root Cause Analysis & DMAIC
13 Reliability Engineering & Maintenance Systems
13.1 Reliability Modeling, Failure Rates (lambda) & Life Distributions (Exponential, Weibull) 13.2 System Reliability Architecture: Series, Parallel, k-out-of-n & Redundancy 13.3 Equipment Maintenance Engineering, Availability (A), MTBF/MTTR & RCM
6.1 Demand Forecasting Methods: Moving Averages, Exponential Smoothing & Error Metrics
Key Takeaways Explain This Topic AI
Demand forecasting methods bifurcate into qualitative techniques (Delphi, market research, historical analogy) for subjective or unquantified environments and quantitative techniques (time series, causal econometric regression) for historical operational data.
Classical time series models decompose observed demand into five structural components: baseline level ( L L L ), secular trend ( T T T ), seasonal variation ( S S S ), cyclical fluctuations ( C C C ), and random noise ( ϵ \epsilon ϵ ).
Simple Moving Averages weight past periods equally, whereas Simple Exponential Smoothing (SES) applies geometrically decaying weights controlled by smoothing parameter α ∈ ( 0 , 1 ) \alpha \in (0, 1) α ∈ ( 0 , 1 ) , where larger α \alpha α values increase responsiveness to shifts at the expense of noise dampening.
Holt's two-parameter linear exponential smoothing decouples the level ( L t L_t L t ) and trend ( T t T_t T t ) updates, resolving the persistent mathematical lag exhibited by Simple Exponential Smoothing when tracking trending demand profiles.
Forecast error tracking requires multiple complementary metrics: Mean Absolute Deviation (MAD), Mean Squared Error (MSE), and Mean Absolute Percentage Error (MAPE); the Tracking Signal ( T S t = R S F E t / M A D t TS_t = RSFE_t / MAD_t T S t = RSF E t / M A D t ) triggers automated recalibration when crossing control bounds of ± 4 \pm 4 ± 4 or ± 5 \pm 5 ± 5 MAD.
Last updated: October 2026
Demand Forecasting Methods: Moving Averages, Exponential Smoothing & Error Metrics
Core Principle: Operational capacity planning, master scheduling, and inventory systems depend fundamentally upon demand forecasting. Industrial engineers must select forecasting architectures that minimize both random variance and systematic bias, establishing statistical tracking mechanisms to trigger model recalibration when process dynamics shift.
Every downstream operation in production and logistics—from aggregate workforce planning and master production scheduling to safety stock sizing and material procurement—originates from a demand forecast. Because forecasts are virtually never 100% accurate, industrial and systems engineers focus on three core objectives: selecting appropriate mathematical forecasting models, quantifying uncertainty, and implementing statistical control mechanisms to detect structural shifts in customer demand.
1. Qualitative vs. Quantitative Forecasting Architectures
Forecasting methodologies are broadly divided into qualitative (judgmental) and quantitative (mathematical) paradigms based on the availability and relevance of historical data.
Dimension Qualitative Methods Quantitative: Causal Models Quantitative: Time Series
Data Requirements Subjective expertise, intuition, consumer surveys Historical response variable and independent explanatory predictors Chronological historical demand sequences ( A t , A t − 1 , … A_t, A_{t-1}, \dots A t , A t − 1 , … )
Primary Methods Delphi technique, panel consensus, market research, historical analogy Ordinary least squares (OLS) regression, multiple linear regression, econometric models Simple moving average (SMA), weighted moving average (WMA), exponential smoothing (SES, Holt, Winter)
Time Horizon Long-range strategic planning (2–10 years), new facility site selection, new product introduction (NPI) Medium-to-long range (1–3 years), macro-economic demand drivers Short-to-medium tactical operations (1 day to 12 months), shop-floor scheduling, MRP
Key Strengths Captures technological breakthroughs, regulatory shifts, and paradigm disruptions Identifies root-cause functional relationships ( Y = f ( X 1 , X 2 ) Y = f(X_1, X_2) Y = f ( X 1 , X 2 ) ) Highly automated, computationally efficient, mathematically objective
Core Limitations Cognitive bias, groupthink, lack of mathematical repeatability Requires forecasting the independent variables; collinearity risks Assumes historical patterns persist; lags sudden structural shifts
The Delphi Method
Among qualitative methods, the Delphi method is prominent in systems engineering. Developed by the RAND Corporation, it achieves objective expert consensus while eliminating the interpersonal biases of group discussions:
Facilitator Panel Selection: A steering coordinator recruits an independent panel of domain experts.
Iterative Questionnaire Rounds: Experts answer anonymous questionnaires regarding technological timelines or market projections.
Statistical Feedback Summary: The facilitator compiles responses into statistical summaries (median, interquartile range) and qualitative rationales, returning them to the panel.
Convergence: Panelists review the anonymized rationales and revise their estimates over multiple rounds (typically 3 to 4 iterations) until opinions converge to a stable consensus.
2. Classical Time Series Decomposition
A chronological sequence of historical demand observations A t A_t A t (where t = 1 , 2 , … , n t = 1, 2, \dots, n t = 1 , 2 , … , n ) can be decomposed into five structural mathematical components:
Level ( L t L_t L t ): The baseline or central tendency of the series if all variations were eliminated.
Trend ( T t T_t T t ): The persistent, long-term upward or downward secular slope over multiple periods.
Seasonality ( S t S_t S t ): Repeating, predictable cyclical patterns linked to calendar intervals (e.g., weekly shifts in parcel processing, quarterly surges in retail manufacturing).
Cyclicality ( C t C_t C t ): Wavelike macroeconomic expansions and contractions that span multi-year durations with varying amplitudes and non-fixed periodicity.
Random Noise ( ϵ t \epsilon_t ϵ t ): Uncorrelated, non-deterministic random variation assumed to follow an independent and identically distributed normal distribution ϵ t ∼ N ( 0 , σ 2 ) \epsilon_t \sim \mathcal{N}(0, \sigma^2) ϵ t ∼ N ( 0 , σ 2 ) . Time Series Composition
Observed Demand A_t = Level (L) + Trend (T) + Seasonality (S) + Cyclical (C) + Noise (ε)
- Demand ^
- | / Combined Time Series
- | /\ /
- | /\ / \ /
- | /\ / \ / \ /
- | /\ / \ / \ / v
- | /\ / \ / \ /
- | / \ / \ /
- | / \ /
- | / ---------------- Secular Linear Trend (T)
- +--------------------------------------------------------> Time (t)
- Additive vs. Multiplicative Models
Additive Decomposition: Assumes seasonal fluctuations possess constant absolute amplitude regardless of the baseline level:
A t = L t + T t + S t + C t + ϵ t A_t = L_t + T_t + S_t + C_t + \epsilon_t A t = L t + T t + S t + C t + ϵ t
Multiplicative Decomposition: Assumes seasonal variation scales proportionally with the baseline demand level:
A t = ( L t + T t ) × S t × C t × ϵ t A_t = (L_t + T_t) \times S_t \times C_t \times \epsilon_t A t = ( L t + T t ) × S t × C t × ϵ t
In industrial supply chains, the multiplicative model is widely applied because seasonal swings typically scale with business growth.
3. Moving Averages & Weighted Moving Averages
When demand exhibits no persistent trend or seasonality, moving averages filter out high-frequency random noise.
Simple Moving Average (SMA)
The n n n -period Simple Moving Average forecasts demand for period t t t by taking the unweighted arithmetic mean of the preceding n n n actual observations:
S M A n ( t ) = 1 n ∑ i = 1 n A t − i = A t − 1 + A t − 2 + ⋯ + A t − n n SMA_n(t) = \frac{1}{n} \sum_{i=1}^n A_{t-i} = \frac{A_{t-1} + A_{t-2} + \dots + A_{t-n}}{n} SM A n ( t ) = n 1 ∑ i = 1 n A t − i = n A t − 1 + A t − 2 + ⋯ + A t − n
The Stability vs. Responsiveness Trade-off:
Small n n n (e.g., n = 3 n = 3 n = 3 ): Rapidly tracks real shifts in demand level, but reacts excessively to transient random noise.
Large n n n (e.g., n = 12 n = 12 n = 12 ): Highly stable and smooths out noise effectively, but severely lags behind genuine shifts in baseline level.
Weighted Moving Average (WMA)
The Weighted Moving Average assigns unequal weights w i w_i w i to historical observations, typically weighting recent periods more heavily:
W M A ( t ) = ∑ i = 1 n w i A t − i = w 1 A t − 1 + w 2 A t − 2 + ⋯ + w n A t − n WMA(t) = \sum_{i=1}^n w_i A_{t-i} = w_1 A_{t-1} + w_2 A_{t-2} + \dots + w_n A_{t-n} W M A ( t ) = ∑ i = 1 n w i A t − i = w 1 A t − 1 + w 2 A t − 2 + ⋯ + w n A t − n
Subject to the normalization condition:
- ∑ i = 1 n w i = 1.0 , w i ≥ 0 \sum_{i=1}^n w_i = 1.0, \quad w_i \ge 0 ∑ i = 1 n w i = 1.0 , w i ≥ 0
- 4. Simple Exponential Smoothing (SES)
Simple Exponential Smoothing (SES) is an autoregressive moving average technique that applies geometrically declining weights to past demand. It requires storing only two values: the prior forecast F t − 1 F_{t-1} F t − 1 and the prior actual demand A t − 1 A_{t-1} A t − 1 .
Mathematical Formulation
The fundamental recurrence relation is:
F t = F t − 1 + α ( A t − 1 − F t − 1 ) = α A t − 1 + ( 1 − α ) F t − 1 F_t = F_{t-1} + \alpha (A_{t-1} - F_{t-1}) = \alpha A_{t-1} + (1 - \alpha) F_{t-1} F t = F t − 1 + α ( A t − 1 − F t − 1 ) = α A t − 1 + ( 1 − α ) F t − 1
Where:
F t F_t F t is the forecast for period t t t .
A t − 1 A_{t-1} A t − 1 is the actual demand observed in period t − 1 t-1 t − 1 .
F t − 1 F_{t-1} F t − 1 is the forecast generated for period t − 1 t-1 t − 1 .
( A t − 1 − F t − 1 ) = e t − 1 (A_{t-1} - F_{t-1}) = e_{t-1} ( A t − 1 − F t − 1 ) = e t − 1 is the forecast error of the preceding period.
α \alpha α is the smoothing constant, bounded by 0 ≤ α ≤ 1 0 \le \alpha \le 1 0 ≤ α ≤ 1 .
Expanding the Geometric Series
Substituting F t − 1 = α A t − 2 + ( 1 − α ) F t − 2 F_{t-1} = \alpha A_{t-2} + (1 - \alpha)F_{t-2} F t − 1 = α A t − 2 + ( 1 − α ) F t − 2 recursively demonstrates how historical weights decay:
F t = α A t − 1 + α ( 1 − α ) A t − 2 + α ( 1 − α ) 2 A t − 3 + ⋯ + ( 1 − α ) k F t − k F_t = \alpha A_{t-1} + \alpha(1 - \alpha) A_{t-2} + \alpha(1 - \alpha)^2 A_{t-3} + \dots + (1 - \alpha)^k F_{t-k} F t = α A t − 1 + α ( 1 − α ) A t − 2 + α ( 1 − α ) 2 A t − 3 + ⋯ + ( 1 − α ) k F t − k
Because ( 1 − α ) < 1 (1 - \alpha) < 1 ( 1 − α ) < 1 , the influence of past observations decreases exponentially with age. Weight Assigned to Historical Observations (Exponential Decay)
- Weight ^
- | * [α]
- | \
- | * [α(1-α)]
- | \
- | * [α(1-α)²]
- | \
- | * [α(1-α)³]
- | \----
- +---------------------------------------------> Lag Periods into Past
- t-1 t-2 t-3 t-4 t-5
- Selection of the Smoothing Constant α \alpha α
The equivalent moving average span n n n that matches the average age of observations under SES is:
n ≈ 2 − α α ⟺ α = 2 n + 1 n \approx \frac{2 - \alpha}{\alpha} \iff \alpha = \frac{2}{n + 1} n ≈ α 2 − α ⟺ α = n + 1 2
Low α \alpha α ( 0.05 ≤ α ≤ 0.15 0.05 \le \alpha \le 0.15 0.05 ≤ α ≤ 0.15 ): Corresponds to a large moving average window ( n ≈ 12 n \approx 12 n ≈ 12 to 39 39 39 ). High filtering stability; ideal for stable products with high random noise.
High α \alpha α ( 0.30 ≤ α ≤ 0.50 0.30 \le \alpha \le 0.50 0.30 ≤ α ≤ 0.50 ): Corresponds to a short window ( n ≈ 3 n \approx 3 n ≈ 3 to 6 6 6 ). Rapid response; ideal for dynamic, low-noise demand environments.
Response to Step Changes: When demand undergoes an instantaneous step change of magnitude Δ \Delta Δ , the SES forecast tracks toward the new level with asymptotic convergence:
F t + k = F t + Δ ( 1 − ( 1 − α ) k ) F_{t+k} = F_t + \Delta \left( 1 - (1 - \alpha)^k \right) F t + k = F t + Δ ( 1 − ( 1 − α ) k )
5. Trend-Adjusted Exponential Smoothing (Holt's Model)
When a time series exhibits a systematic trend, Simple Exponential Smoothing exhibits a perpetual lag : it consistently underestimates demand during an upward trend and overestimates demand during a downward trend. The steady-state lag under a linear trend with slope T T T is:
Lag S E S = 1 − α α T \text{Lag}_{SES} = \frac{1 - \alpha}{\alpha} T Lag SES = α 1 − α T
To eliminate this lag, Charles C. Holt introduced two-parameter linear exponential smoothing , which separates the baseline level from the trend component.
Holt's Mathematical Formulation
At the conclusion of period t t t , upon observing actual demand A t A_t A t , two smoothing equations update the state variables:
Level Equation ( L t L_t L t ):
L t = α A t + ( 1 − α ) ( L t − 1 + T t − 1 ) L_t = \alpha A_t + (1 - \alpha)(L_{t-1} + T_{t-1}) L t = α A t + ( 1 − α ) ( L t − 1 + T t − 1 )
Trend Equation ( T t T_t T t ):
T t = β ( L t − L t − 1 ) + ( 1 − β ) T t − 1 T_t = \beta (L_t - L_{t-1}) + (1 - \beta) T_{t-1} T t = β ( L t − L t − 1 ) + ( 1 − β ) T t − 1
Forecast Generation for k k k Periods Ahead ( F t + k F_{t+k} F t + k ):
F t + k = L t + k T t F_{t+k} = L_t + k T_t F t + k = L t + k T t
Where:
L t L_t L t is the smoothed level at period t t t .
T t T_t T t is the smoothed trend (slope per period) at period t t t .
α \alpha α is the level smoothing parameter ( 0 < α < 1 0 < \alpha < 1 0 < α < 1 ).
β \beta β is the trend smoothing parameter ( 0 < β < 1 0 < \beta < 1 0 < β < 1 ).
k k k is the forecast lead time horizon ( k ≥ 1 k \ge 1 k ≥ 1 ).
Seasonal Decomposition and Winter's Model Overview
For series exhibiting both trend and seasonality, Winter's Three-Parameter Model incorporates a seasonal multiplicative factor S t S_t S t :
L t = α ( A t S t − p ) + ( 1 − α ) ( L t − 1 + T t − 1 ) L_t = \alpha \left(\frac{A_t}{S_{t-p}}\right) + (1 - \alpha)(L_{t-1} + T_{t-1}) L t = α ( S t − p A t ) + ( 1 − α ) ( L t − 1 + T t − 1 ) T t = β ( L t − L t − 1 ) + ( 1 − β ) T t − 1 T_t = \beta (L_t - L_{t-1}) + (1 - \beta) T_{t-1} T t = β ( L t − L t − 1 ) + ( 1 − β ) T t − 1 S t = γ ( A t L t ) + ( 1 − γ ) S t − p S_t = \gamma \left(\frac{A_t}{L_t}\right) + (1 - \gamma) S_{t-p} S t = γ ( L t A t ) + ( 1 − γ ) S t − p F t + k = ( L t + k T t ) × S t − p + k F_{t+k} = (L_t + k T_t) \times S_{t-p+k} F t + k = ( L t + k T t ) × S t − p + k
Where p p p is the seasonal period (e.g., p = 4 p = 4 p = 4 for quarters, p = 12 p = 12 p = 12 for months), and γ \gamma γ is the seasonal smoothing parameter.
6. Forecast Error Metrics and Bias Tracking
The forecast error in period t t t is defined as:
e t = A t − F t e_t = A_t - F_t e t = A t − F t
Evaluating forecast models requires measuring both dispersion (magnitude of error) and bias (directional skew). Error Metric Hierarchy
- Forecast Error e_t = Actual (A_t) - Forecast (F_t)
- ├── Dispersion Metrics (Error Magnitude)
- │ ├── MAD = (1/n) Σ |e_t| [Linear penalty; robust to outliers; σ ≈ 1.25 MAD]
- │ ├── MSE = (1/n) Σ (e_t)² [Quadratic penalty; heavily penalizes large errors; σ ≈ √MSE]
- │ └── MAPE = (1/n) Σ (|e_t|/A_t) × 100% [Scale-independent percentage; enables cross-SKU comparisons]
- └── Bias Metrics (Directional Skew)
- ├── Mean Error (ME) = (1/n) Σ e_t [Detects persistent over/under-forecasting]
- ├── RSFE = Σ e_t [Cumulative sum of errors]
- └── Tracking Signal (TS) = RSFE / MAD [Statistical control limit: ±4 to ±5 MAD]
- Detailed Mathematical Definitions
Mean Absolute Deviation (MAD): Measures the average absolute error magnitude:
M A D = 1 n ∑ t = 1 n ∣ A t − F t ∣ = 1 n ∑ t = 1 n ∣ e t ∣ MAD = \frac{1}{n} \sum_{t=1}^n |A_t - F_t| = \frac{1}{n} \sum_{t=1}^n |e_t| M A D = n 1 ∑ t = 1 n ∣ A t − F t ∣ = n 1 ∑ t = 1 n ∣ e t ∣
For normally distributed errors with zero mean, the standard deviation of demand σ \sigma σ is related to MAD by:
σ ≈ π 2 × M A D ≈ 1.2533 × M A D ⟺ M A D ≈ 0.7979 × σ \sigma \approx \sqrt{\frac{\pi}{2}} \times MAD \approx 1.2533 \times MAD \iff MAD \approx 0.7979 \times \sigma σ ≈ 2 π × M A D ≈ 1.2533 × M A D ⟺ M A D ≈ 0.7979 × σ
Mean Squared Error (MSE): Measures variance by penalizing large outlier errors quadratically:
M S E = 1 n ∑ t = 1 n ( A t − F t ) 2 = 1 n ∑ t = 1 n e t 2 MSE = \frac{1}{n} \sum_{t=1}^n (A_t - F_t)^2 = \frac{1}{n} \sum_{t=1}^n e_t^2 MSE = n 1 ∑ t = 1 n ( A t − F t ) 2 = n 1 ∑ t = 1 n e t 2 σ ≈ M S E \sigma \approx \sqrt{MSE} σ ≈ MSE
Mean Absolute Percentage Error (MAPE): Evaluates relative error, normalizing by actual demand volume to allow comparison across high-volume and low-volume products:
M A P E = 1 n ∑ t = 1 n ∣ A t − F t ∣ A t × 100 % MAPE = \frac{1}{n} \sum_{t=1}^n \frac{|A_t - F_t|}{A_t} \times 100\% M A PE = n 1 ∑ t = 1 n A t ∣ A t − F t ∣ × 100%
Tracking Signal ( T S TS TS ) and Statistical Control Limits
A model may have a low MAD but still suffer from severe systematic bias (e.g., consistently under-forecasting demand). The Running Sum of Forecast Errors ( R S F E RSFE RSFE ) , also called cumulative forecast error, tracks bias accumulation:
R S F E t = ∑ i = 1 t e i = ∑ i = 1 t ( A i − F i ) RSFE_t = \sum_{i=1}^t e_i = \sum_{i=1}^t (A_i - F_i) RSF E t = ∑ i = 1 t e i = ∑ i = 1 t ( A i − F i )
The Tracking Signal ( T S t TS_t T S t ) normalizes R S F E RSFE RSFE by the contemporary Mean Absolute Deviation:
T S t = R S F E t M A D t = ∑ i = 1 t ( A i − F i ) 1 t ∑ i = 1 t ∣ A i − F i ∣ TS_t = \frac{RSFE_t}{MAD_t} = \frac{\sum_{i=1}^t (A_i - F_i)}{\frac{1}{t} \sum_{i=1}^t |A_i - F_i|} T S t = M A D t RSF E t = t 1 ∑ i = 1 t ∣ A i − F i ∣ ∑ i = 1 t ( A i − F i )
Interpretation:
T S t = 0 TS_t = 0 T S t = 0 : Perfect unbiased calibration.
T S t > 0 TS_t > 0 T S t > 0 : Actual demand consistently exceeds forecast (under-forecasting; risks stockouts).
T S t < 0 TS_t < 0 T S t < 0 : Actual demand consistently falls below forecast (over-forecasting; causes excess inventory).
Control Limits: Operational control bounds are typically set at ± 4.0 \pm 4.0 ± 4.0 to ± 5.0 \pm 5.0 ± 5.0 MAD . If ∣ T S t ∣ > 4.0 |TS_t| > 4.0 ∣ T S t ∣ > 4.0 , the probability that the error is due solely to random variation is under 3% (equivalent to approximately ± 3 σ \pm 3\sigma ± 3 σ ). Exceeding these bounds signals structural change in demand or model obsolescence, requiring immediate retuning of parameters or switching models.
7. Comprehensive Worked Numerical Example
A precision manufacturing plant tracks monthly orders for an industrial gearbox. The actual demands observed over a 6-month period are:
- Period 1: A 1 = 120 A_1 = 120 A 1 = 120 units
- Period 2: A 2 = 135 A_2 = 135 A 2 = 135 units
- Period 3: A 3 = 130 A_3 = 130 A 3 = 130 units
- Period 4: A 4 = 150 A_4 = 150 A 4 = 150 units
- Period 5: A 5 = 165 A_5 = 165 A 5 = 165 units
- Period 6: A 6 = 160 A_6 = 160 A 6 = 160 units
We evaluate and compare three forecasting models:
Model A: 3-period Simple Moving Average ( S M A 3 SMA_3 SM A 3 ).
Model B: 3-period Weighted Moving Average ( W M A 3 WMA_3 W M A 3 ) with weights w 1 = 0.50 w_1 = 0.50 w 1 = 0.50 (lag 1), w 2 = 0.30 w_2 = 0.30 w 2 = 0.30 (lag 2), and w 3 = 0.20 w_3 = 0.20 w 3 = 0.20 (lag 3).
Model C: Simple Exponential Smoothing ( S E S SES SES ) with α = 0.30 \alpha = 0.30 α = 0.30 , initialized with F 1 = 120.0 F_1 = 120.0 F 1 = 120.0 .
Step-by-Step Forecast Generation
1. Calculating S M A 3 SMA_3 SM A 3 :
F 4 = A 3 + A 2 + A 1 3 = 130 + 135 + 120 3 = 385 3 = 128.33 F_4 = \frac{A_3 + A_2 + A_1}{3} = \frac{130 + 135 + 120}{3} = \frac{385}{3} = 128.33 F 4 = 3 A 3 + A 2 + A 1 = 3 130 + 135 + 120 = 3 385 = 128.33
F 5 = A 4 + A 3 + A 2 3 = 150 + 130 + 135 3 = 415 3 = 138.33 F_5 = \frac{A_4 + A_3 + A_2}{3} = \frac{150 + 130 + 135}{3} = \frac{415}{3} = 138.33 F 5 = 3 A 4 + A 3 + A 2 = 3 150 + 130 + 135 = 3 415 = 138.33
F 6 = A 5 + A 4 + A 3 3 = 165 + 150 + 130 3 = 445 3 = 148.33 F_6 = \frac{A_5 + A_4 + A_3}{3} = \frac{165 + 150 + 130}{3} = \frac{445}{3} = 148.33 F 6 = 3 A 5 + A 4 + A 3 = 3 165 + 150 + 130 = 3 445 = 148.33
Forecast for Period 7: F 7 = A 6 + A 5 + A 4 3 = 160 + 165 + 150 3 = 158.33 F_7 = \frac{A_6 + A_5 + A_4}{3} = \frac{160 + 165 + 150}{3} = 158.33 F 7 = 3 A 6 + A 5 + A 4 = 3 160 + 165 + 150 = 158.33
2. Calculating W M A 3 WMA_3 W M A 3 :
F 4 = 0.50 ( 130 ) + 0.30 ( 135 ) + 0.20 ( 120 ) = 65.0 + 40.5 + 24.0 = 129.50 F_4 = 0.50(130) + 0.30(135) + 0.20(120) = 65.0 + 40.5 + 24.0 = 129.50 F 4 = 0.50 ( 130 ) + 0.30 ( 135 ) + 0.20 ( 120 ) = 65.0 + 40.5 + 24.0 = 129.50
F 5 = 0.50 ( 150 ) + 0.30 ( 130 ) + 0.20 ( 135 ) = 75.0 + 39.0 + 27.0 = 141.00 F_5 = 0.50(150) + 0.30(130) + 0.20(135) = 75.0 + 39.0 + 27.0 = 141.00 F 5 = 0.50 ( 150 ) + 0.30 ( 130 ) + 0.20 ( 135 ) = 75.0 + 39.0 + 27.0 = 141.00
F 6 = 0.50 ( 165 ) + 0.30 ( 150 ) + 0.20 ( 130 ) = 82.5 + 45.0 + 26.0 = 153.50 F_6 = 0.50(165) + 0.30(150) + 0.20(130) = 82.5 + 45.0 + 26.0 = 153.50 F 6 = 0.50 ( 165 ) + 0.30 ( 150 ) + 0.20 ( 130 ) = 82.5 + 45.0 + 26.0 = 153.50
Forecast for Period 7: F 7 = 0.50 ( 160 ) + 0.30 ( 165 ) + 0.20 ( 150 ) = 80.0 + 49.5 + 30.0 = 159.50 F_7 = 0.50(160) + 0.30(165) + 0.20(150) = 80.0 + 49.5 + 30.0 = 159.50 F 7 = 0.50 ( 160 ) + 0.30 ( 165 ) + 0.20 ( 150 ) = 80.0 + 49.5 + 30.0 = 159.50
3. Calculating S E S SES SES ( α = 0.30 \alpha = 0.30 α = 0.30 ):
F 1 = 120.00 F_1 = 120.00 F 1 = 120.00
F 2 = 120.00 + 0.30 ( 120 − 120.00 ) = 120.00 F_2 = 120.00 + 0.30(120 - 120.00) = 120.00 F 2 = 120.00 + 0.30 ( 120 − 120.00 ) = 120.00
F 3 = 120.00 + 0.30 ( 135 − 120.00 ) = 120.00 + 4.50 = 124.50 F_3 = 120.00 + 0.30(135 - 120.00) = 120.00 + 4.50 = 124.50 F 3 = 120.00 + 0.30 ( 135 − 120.00 ) = 120.00 + 4.50 = 124.50
F 4 = 124.50 + 0.30 ( 130 − 124.50 ) = 124.50 + 1.65 = 126.15 F_4 = 124.50 + 0.30(130 - 124.50) = 124.50 + 1.65 = 126.15 F 4 = 124.50 + 0.30 ( 130 − 124.50 ) = 124.50 + 1.65 = 126.15
F 5 = 126.15 + 0.30 ( 150 − 126.15 ) = 126.15 + 7.155 = 133.31 F_5 = 126.15 + 0.30(150 - 126.15) = 126.15 + 7.155 = 133.31 F 5 = 126.15 + 0.30 ( 150 − 126.15 ) = 126.15 + 7.155 = 133.31
F 6 = 133.31 + 0.30 ( 165 − 133.31 ) = 133.31 + 9.507 = 142.82 F_6 = 133.31 + 0.30(165 - 133.31) = 133.31 + 9.507 = 142.82 F 6 = 133.31 + 0.30 ( 165 − 133.31 ) = 133.31 + 9.507 = 142.82
Forecast for Period 7: F 7 = 142.82 + 0.30 ( 160 − 142.82 ) = 142.82 + 5.154 = 147.97 F_7 = 142.82 + 0.30(160 - 142.82) = 142.82 + 5.154 = 147.97 F 7 = 142.82 + 0.30 ( 160 − 142.82 ) = 142.82 + 5.154 = 147.97
Comparative Forecast and Error Ledger Table
Period ( t t t ) Actual ( A t A_t A t ) F t F_t F t ( S M A 3 SMA_3 SM A 3 ) F t F_t F t ( W M A 3 WMA_3 W M A 3 ) F t F_t F t ( S E S 0.3 SES_{0.3} SE S 0.3 ) e t e_t e t ( S E S SES SES ) ∣ e t ∣ \lvert e_t \rvert ∣ e t ∣ e t 2 e_t^2 e t 2 ∣ e t ∣ / A t \lvert e_t \rvert / A_t ∣ e t ∣ / A t R S F E t RSFE_t RSF E t M A D t MAD_t M A D t T S t TS_t T S t
- 1 120 — — 120.00 0.00 0.00 0.00 0.00% 0.00 0.00 0.00
- 2 135 — — 120.00 +15.00 15.00 225.00 11.11% +15.00 7.50 +2.00
- 3 130 — — 124.50 +5.50 5.50 30.25 4.23% +20.50 6.83 +3.00
- 4 150 128.33 129.50 126.15 +23.85 23.85 568.82 15.90% +44.35 11.09 +4.00
- 5 165 138.33 141.00 133.31 +31.69 31.69 1004.26 19.21% +76.04 15.21 +5.00
- 6 160 148.33 153.50 142.82 +17.18 17.18 295.15 10.74% +93.22 15.54 +6.00
- Analysis and Engineering Findings
Evaluation of SES Error Metrics Across All 6 Periods: M A D = 0 + 15 + 5.5 + 23.85 + 31.69 + 17.18 6 = 93.22 6 = 15.54 units MAD = \frac{0 + 15 + 5.5 + 23.85 + 31.69 + 17.18}{6} = \frac{93.22}{6} = 15.54 \text{ units} M A D = 6 0 + 15 + 5.5 + 23.85 + 31.69 + 17.18 = 6 93.22 = 15.54 units M S E = 0 + 225 + 30.25 + 568.82 + 1004.26 + 295.15 6 = 2123.48 6 = 353.91 units 2 MSE = \frac{0 + 225 + 30.25 + 568.82 + 1004.26 + 295.15}{6} = \frac{2123.48}{6} = 353.91 \text{ units}^2 MSE = 6 0 + 225 + 30.25 + 568.82 + 1004.26 + 295.15 = 6 2123.48 = 353.91 units 2 RMSE = 353.91 = 18.81 units \text{RMSE} = \sqrt{353.91} = 18.81 \text{ units} RMSE = 353.91 = 18.81 units M A P E = 0.00 % + 11.11 % + 4.23 % + 15.90 % + 19.21 % + 10.74 % 6 = 61.19 % 6 = 10.20 % MAPE = \frac{0.00\% + 11.11\% + 4.23\% + 15.90\% + 19.21\% + 10.74\%}{6} = \frac{61.19\%}{6} = 10.20\% M A PE = 6 0.00% + 11.11% + 4.23% + 15.90% + 19.21% + 10.74% = 6 61.19% = 10.20%
Tracking Signal Analysis: In Period 6, R S F E 6 = + 93.22 RSFE_6 = +93.22 RSF E 6 = + 93.22 and M A D 6 = 15.54 MAD_6 = 15.54 M A D 6 = 15.54 , yielding a Tracking Signal of: T S 6 = + 93.22 15.54 = + 6.00 TS_6 = \frac{+93.22}{15.54} = +6.00 T S 6 = 15.54 + 93.22 = + 6.00 Because T S 6 = + 6.00 > + 4.0 TS_6 = +6.00 > +4.0 T S 6 = + 6.00 > + 4.0 , the SES model has breached the upper control limit. Every single forecast error from Period 2 onward was strictly positive ( e t > 0 e_t > 0 e t > 0 ), proving that actual demand grew systematically while the SES model lagged behind. The plant must transition from Simple Exponential Smoothing to Holt's trend-adjusted model to account for the secular growth rate of approximately + 8 +8 + 8 to + 10 +10 + 10 units/month.
8. Causal Forecasting with Simple Linear Regression
Time series methods use only past demand. A causal model forecasts demand from a leading indicator, such as housing starts for a window manufacturer or flight schedules for an aircraft parts supplier. The simplest causal model is a least-squares line y ^ = a + b x \hat{y} = a + bx y ^ = a + b x :
b = n ∑ x y − ∑ x ∑ y n ∑ x 2 − ( ∑ x ) 2 , a = y ˉ − b x ˉ b = \frac{n\sum xy - \sum x \sum y}{n\sum x^2 - \left(\sum x\right)^2}, \qquad a = \bar{y} - b\bar{x} b = n ∑ x 2 − ( ∑ x ) 2 n ∑ x y − ∑ x ∑ y , a = y ˉ − b x ˉ
r = n ∑ x y − ∑ x ∑ y [ n ∑ x 2 − ( ∑ x ) 2 ] [ n ∑ y 2 − ( ∑ y ) 2 ] r = \frac{n\sum xy - \sum x \sum y}{\sqrt{\left[n\sum x^2 - \left(\sum x\right)^2\right]\left[n\sum y^2 - \left(\sum y\right)^2\right]}} r = [ n ∑ x 2 − ( ∑ x ) 2 ] [ n ∑ y 2 − ( ∑ y ) 2 ] n ∑ x y − ∑ x ∑ y
The coefficient of determination r 2 r^2 r 2 is the fraction of the variation in y y y explained by x x x .
Worked example: A window plant records regional housing starts x x x (thousands) and its window orders y y y (hundreds of units) for five quarters:
- Quarter x x x y y y x y xy x y x 2 x^2 x 2
- 1 10 230 2,300 100
- 2 12 240 2,880 144
- 3 15 300 4,500 225
- 4 18 315 5,670 324
- 5 20 365 7,300 400
- Sum 75 1,450 22,650 1,193
b = 5 ( 22,650 ) − 75 ( 1,450 ) 5 ( 1,193 ) − 75 2 = 4,500 340 = 13.24 , a = 290 − 13.24 ( 15 ) = 91.5 b = \frac{5(22{,}650) - 75(1{,}450)}{5(1{,}193) - 75^2} = \frac{4{,}500}{340} = 13.24, \qquad a = 290 - 13.24(15) = 91.5 b = 5 ( 1 , 193 ) − 7 5 2 5 ( 22 , 650 ) − 75 ( 1 , 450 ) = 340 4 , 500 = 13.24 , a = 290 − 13.24 ( 15 ) = 91.5
With ∑ y 2 = 432,950 \sum y^2 = 432{,}950 ∑ y 2 = 432 , 950 , the correlation is r = 0.978 r = 0.978 r = 0.978 and r 2 = 0.957 r^2 = 0.957 r 2 = 0.957 , so housing starts explain about 96% of the variation in orders. If next quarter's housing starts are forecast at 22 thousand, the order forecast is 91.5 + 13.24 ( 22 ) ≈ 382.6 91.5 + 13.24(22) \approx 382.6 91.5 + 13.24 ( 22 ) ≈ 382.6 hundred units.
Warning
A causal forecast is only as good as the forecast of its driver. Extrapolating far beyond the observed range of x x x (here 10 to 20 thousand starts) is risky, and a high r r r does not prove cause and effect. Use the calculator's two-variable statistics mode to get a a a , b b b , and r r r quickly on the exam.