%%{init: {"themeVariables": {"fontSize": "24px"}}}%%
flowchart LR
P[Past Data] --> E[Extract Patterns]
E --> F[Forecast + Uncertainty]
F --> C[Decision Costs]
C --> D[Operational or Policy Decision]
D --> L[Learn and Update]
L --> E
From patterns in data \(\rightarrow\) real-world decisions
It is October. You must decide for next summer:
What information do you wish you had?
Expected visitation.
Forecasting reduces uncertainty about the future.
Forecasting is:
A transparent, testable prediction about future values based on past patterns.
Forecasting is not:
Forecasting is:
Operations
Finance
Public policy
A forecast is useful only relative to the decision it informs.
Two models can have similar accuracy, but:
Those create different costs.
Forecast evaluation must consider decision costs.
Every forecast must answer:
Time series are easier to forecast when:
Forecasting is harder with:
Benchmarks are simple forecasts that set a baseline for performance.
Can you beat the guess: “this month-year will look like the same month last year”?
Seasonal naive is strong because:
Never evaluate a forecast on data used to estimate it.
Without out-of-sample evaluation, you reward overfitting.
Not just low error.
A good forecast is:
Different horizons imply different model choices.
Short horizon:
Long horizon:
Would you use the same model for:
Arches reservation policy:
Key modeling question:
Should you assume:
Forecasting involves causal reasoning, not just extrapolation.
Point forecast \(\neq\) full forecast.
Decision-makers need:
Applications:
Suppose July error is 50,000 visitors.
What does that mean economically?
MAE and RMSE are proxies for decision cost.
Forecasts fail when:
Most models assume the future resembles the past.
That assumption is fragile.
Most economic time series combine:
Decomposition helps by:
\[ y_t = T_t + S_t + R_t \]
Forecasting is not: fit once, done.
Forecasting is:
Especially important when policy regimes change.
In this course pipeline:
Forecasting is the bridge between data and planning.
If Arches removes reservations in 2026, how would you forecast 2027 visitation?
Possible strategies:
Scenario: “Visitation is expected to be 15% higher than last year.”
Before hiring 20 more seasonal staff, what else do you need?
Forecasts support decisions under uncertainty.
Forecasting is not about predicting perfectly.
Forecasting is about:
%%{init: {"themeVariables": {"fontSize": "24px"}}}%%
flowchart LR
P[Past Data] --> E[Extract Patterns]
E --> F[Forecast + Uncertainty]
F --> C[Decision Costs]
C --> D[Operational or Policy Decision]
D --> L[Learn and Update]
L --> E