Auto TSB Intermittent

Teunter–Syntetos–Babai (TSB) is a forecasting method designed specifically for intermittent demand, items with sporadic, irregular demand patterns that include many periods with zero demand. TSB is a variation of Croston’s method that avoids the bias toward over forecasting.

Why use TSB

  • Unbiased forecasts, correct the mathematical bias found in Croston’s method.
  • Separate smoothing, uses different parameters (α, β) for demand size and demand timing.
  • More accurate results, produce forecasts closer to actual average demand.
  • Better inventory decisions, reduce overstock and maintain service levels.

The core strength of TSB is automatic adaptation to changing demand patterns without manual intervention. Exponential smoothing continuously updates demand size and demand interval. This approach makes TSB suitable for dynamic environments where demand frequency changes over time. Forecasts are based on recent behavior.

Advantages of TSB

  • Exponential smoothing, recent data has more influence.
  • Separation of size and timing, captures the structure of intermittent demand patterns.
  • Automatic adaptation, responds to changes in demand frequency.
  • Frequency agnostic behavior, does not depend on whether data is weekly, monthly, quarterly, or other.

It is best suited for spare parts and service parts, slow moving inventory items, products with infrequent, unpredictable orders, and items with many zero demand periods.

It is not recommended for regular, consistent demand patterns and trending or seasonal demand.

TSB forecast output formats

TSB provides two optional forecast output formats:

  • Standard forecast (default), expected average demand per period.
  • Intermittent pattern forecast, full demand at expected intervals, with zero demand in other periods.

Mathematical model

TSB separates intermittent demand into three components:

  • Demand size (D) - quantity ordered when demand occurs.
  • Demand interval (I) - average number of periods between orders.
  • Demand probability (P) - likelihood of demand in any period, where (P = 1 / I).

Forecast formula:

Standard forecast = P × D = (1 / I) × D

Example calculation:

  • Historical demand pattern, 60 units every 3 periods.
  • D = 60 (demand size)
  • I = 3 (interval)
  • P = 1 / 3 = 0.333 (probability)
  • Forecast = 0.333 × 60 = 20 units per period

Smoothing parameters

TSB uses exponential smoothing with two parameters, which are automatically optimized:

  • α (alpha), controls smoothing of demand size (typical range from 0.05 to 0.30).
    • Lower values, provide more stability and less responsiveness.
    • Higher values, provide more responsiveness to recent changes.
  • β (beta), controls smoothing of demand interval (typical range from 0.05 to 0.30).
    • Lower values, provide more stable interval estimates.
    • Higher values, provide faster adaptation to changing demand patterns.

The system tests 36 combinations and selects the optimal α and β values. The system selects the pair that minimizes forecast error.