Module3 Flashcards
Demand Forecasting
- Definition: Predicting future demand based on historical data
- Importance: Helps plan production, inventory, sourcing, pricing
- Focus: We study demand forecasting for supply chains
The Role of Forecasting in the Supply Chain
- Basis for all planning
- Affects production, inventory, sales, and new product development
- Involves interrelated processes across the firm
Forecasting Horizons
- Strategic (Long-range): 3+ years
- Tactical (Medium-range): 3 months to 3 years
- Operational (Short-range): Up to 1 year (generally < 3 months)
Demand Forecasting is …
> “Calculating or predicting future demand of a product or service, usually as a result of study and analysis of available historical data.”
Basic Approach to Demand Forecasting
- Understand the objective
- Integrate demand planning across the supply chain
- Identify factors influencing demand
- Forecast at the correct level of aggregation
- Establish performance/error measures
Consequences of Bad Forecasting
- Financial Impact: Over-forecast → Excess inventory; Under-forecast → Lost sales
- Misallocation of resources: Inventory, facilities, transportation, pricing
- Leads to wasted costs or missed opportunities
Qualitative vs. Quantitative Forecasting
- Qualitative: Delphi, Cross-Impact, Historical Analogy
- Quantitative: Causal, Time-Series, Simulation
- Combination often yields more accurate results
Causal Models
- Idea: Demand = f (other external factors)
- Linear Example: Y = b0 + b1X1 + … + bnXn
- Challenge: Identifying correct causal factors
Time-Series Models
- Observed Demand (O) = Systematic (S) + Random (R)
- Systematic can include level, trend, seasonality
- Goal: Identify patterns from historical demand
Forecast Error
- Formula: et = Ft – Dt
- Interpretation: Difference between forecast and actual demand
- High error → Inaccurate forecasts
Moving Average (Time-Series)
- Assumption: No trend or seasonality
- Formula: Lt = (Dt + … + Dt–N+1)/N
- Pros: Simple, stable
- Cons: Lags a trend, ignores complex data relationships
Simple Exponential Smoothing
do we need to know this?
- Use: No trend or seasonality
- Formula: Lt+1 = αDt+1 + (1–α)Lt
- α ∈ [0,1]: Smoothing constant
Parameter α (Alpha)
(in smoothing), I’m not sure if we need to know this
- Small α: Heavier smoothing, slower response to changes
- Large α: Reacts faster, less smoothing
- Choose based on data stability
Holt’s Model (Trend-Corrected Exp. Smoothing)
idk if I’m learning the formulas
- Appropriate: Demand with level + trend
- Forecast: Ft+n = Lt + nTt
-
Update:
- Lt+1 = αDt+1 + (1–α)(Lt + Tt)
- Tt+1 = β(Lt+1 – Lt) + (1–β)Tt
Winter’s Model (Trend + Seasonality)
- Appropriate: Demand with level, trend, seasonality
- Forecast: Ft+n = (Lt + nTt) · St+n
- Updates level, trend, seasonal factor each period
Overview of Adaptive Time-Series (Models for Forecasting)
- No trend/seasonality → Moving Average / Simple Exponential
- Trend → Holt’s Model
- Trend + Seasonality → Winter’s Model
Big Data Characteristics
- Volume: Huge amounts of data
- Variety: Different formats (text, images, etc.)
- Velocity: Rapid generation of data
- Veracity: Uncertainty in data quality
Artificial Intelligence (AI)
- Uses advanced algorithms to interpret events and automate decisions
- Machine learning for real-time adaptation
- Simulates human decision-making steps
AI in Demand Forecasting
- Incorporates massive datasets (ERP, CRM, social media)
- Real-time adaptation to disruptions or new trends
- Focus on strategic decisions rather than manual tasks
Key Learnings
in demand forecasting
- Forecasts are usually wrong
- Provide a range, not a single value
- Short-term forecasts tend to be more accurate
- Aggregation improves accuracy
- Be aware of the bullwhip effect: distortions up the chain
