Forecasting Notes

Chapter 5: Forecasting - In-Depth Notes

Introduction to Forecasting

  • The primary goal of forecasting is to reduce uncertainty and provide better estimates of future occurrences.

Types of Forecasting Models

  • Qualitative Models: Decision-making relies on judgment rather than quantifiable data. Suitable when accurate data is hard to obtain.

    • Common Techniques:

    • Delphi Method: Involves experts who provide feedback through multiple rounds to reach consensus. decision-makers staff personnel and respondents.

    • Jury of Executive Opinion: Utilizes a panel of high-level managers to form a collective estimate. resulting group estimates.

    • Sales Force Composite: Collects projections from individual salespersons, aggregated for a broader forecast.

    • Consumer Market Surveys: Gathers customer purchase intentions to inform forecasts.

  • Time-Series Methods: Uses historical data to predict future values based on past trends.

    • Components of Time Series:

    • Trend (T): Long-term increase or decrease.

    • Seasonal (S): Regular fluctuations during specific periods.

    • Cyclical (C): Patterns that occur over longer durations, typically years.

    • Random (R): Unpredictable variations.

    • Basic Forms: Can be multiplicative (T × S × C × R) or additive (T + S + C + R).

  • Causal Methods: Establish relationships between dependent and independent variables, typically using regression analysis.

Measures of Forecast Accuracy

  • Accuracy is determined by comparing forecasts with actual outcomes.

  • Common Accuracy Metrics:

    • Mean Absolute Deviation (MAD): Average of absolute differences between actual and forecasted values.

    • Mean Squared Error (MSE): Average of squared errors, emphasizing larger errors more than smaller ones.

    • Mean Absolute Percent Error (MAPE): Evaluates forecast accuracy as a percentage.

Forecasting Models for Random Variations Only

  • Forecasts that only consider random variations are simplified:

    • Averaging Techniques:

    • Moving Averages: Averages over a specified number of periods to smooth out short-term fluctuations.

    • Weighted Moving Averages: Assigns different weights to past observations, giving more importance to recent data.

    • Exponential Smoothing: Similar to moving averages, but places emphasis on recent observations more systematically using a smoothing constant (α).

Forecasting Models Considering Trend and Random Variations

  • Exponential Smoothing with Trend: An extension of basic exponential smoothing that adjusts forecasts for trends using an additional smoothing constant (β).

    • Simple Steps:

    1. Compute a smoothed forecast combining previous forecast and errors.

    2. Update trend using smoothed forecasts.

    3. Calculate final forecast including the trend adjustment.

  • Trend Projection Models: Fit a regression line to historical data, predicting future values based on established trends.

Seasonal Variations

  • Seasonality: Regular patterns in data at specific periods that require adjustments.

  • Seasonal indices can indicate deviations from average performance during specific times (e.g., index > 1 indicates a stronger season).

    • Procedures to calculate:

    1. Create seasonal indices by averaging over multiple periods.

    2. Deseasonalize data by dividing observed sales by their respective seasonal index.

    3. Re-adjust forecasts by applying seasonal indices.

Combining All Components (Trend, Seasonal, Random)

  • Decomposition Method: Isolates linear trends and seasonal factors for accurate forecasts.

    • Steps:

    1. Compute seasonal indices.

    2. Deseasonalize data.

    3. Develop a trend line from deseasonalized data.

    4. Forecast future data with the trend.

    5. Adjust forecasts with seasonal indices.

  • Multiple Linear Regression: Utilizes both time as an independent variable and dummy variables to represent seasonal trends simultaneously to predict outcomes effectively.

Monitoring and Controlling Forecasts

  • Tracking Signal: Evaluates forecast accuracy by comparing cumulative forecast errors to MAD. Alerts to persistent deviations.

    • Positive signals suggest demand exceeds forecast, negative indicates shortfalls.

Example Applications

  • Adaptive Smoothing: Adjusts keys parameters (α, β) in real-time based on forecast errors to optimize future forecasting efforts.

  • Excel Implementation: Steps to implement various forecasting models in Excel for practical understanding and applications in forecasting scenarios.