267. Smoothing
Operators that compute smoothed (averaged) versions of input signals. Implemented as exponential delays but used to model perception rather than physical transit.
267.1. First-order smoothing (exponential smoothing)
tracks with smoothing time — equivalent to single-exponential smoothing.
In discrete time: with .
In Vensim: SMOOTH(input, tau).
267.2. Higher-order smoothing
first-order smooths in series — each one smoother. SMOOTH3 is third-order (very common default in SD models). Removes more noise but adds more phase lag.
267.3. TREND function
Computes a smoothed fractional rate of change:
where is smoothed with time . Gives “perceived growth rate” — used to model planners’ expectations.
In Vensim: TREND(input, perception_time, trend_time).
267.4. FORECAST function
Extrapolates the input forward using TREND:
Forecast time units ahead based on current trend.
Used to model planning under expectations — what someone thinks future demand will be.
267.5. Why smoothing models decision-making
People (and organizations) don’t react to raw data; they react to smoothed versions:
- Expected demand: smoothed over weeks / months
- Perceived inflation: long-term average, not daily news
- Expected supply lead time: averaged over past orders
Smoothing operators in SD capture this adaptive expectation behavior. Combined with delays and stocks, they generate the “policy resistance” and stability problems that define Sterman’s beer-game style dynamics.
267.6. Phase lag
Smoothing introduces a delay in perception:
- First-order smooth: lag is approximately
- Third-order smooth: lag approximately but with much sharper response
- Higher orders: tighter pulse, slower mean response
Phase lag = part of why supply-chain bullwhip happens: each echelon’s reaction is delayed relative to actual demand.
267.7. See also
- Delays — same mechanism, different framing
- Stock Management — uses smoothing for expected demand
- Exponential Smoothing (ETS) — discrete-time analog