Section 14 of 24 · 6 lessons

Time Series and Forecasting

Tomorrow's demand is unknown when Botie needs to prepare for it. Start with a forecast that simply repeats a known value, then introduce earlier observations, repeating patterns, and fitted relationships. Every comparison will respect what could have been known at the time the forecast was made.

0 of 6 finished

Predicting the Next Observation

Tomorrow's value is unavailable when we make today's forecast. Begin with a simple prediction, then arrange earlier observations into inputs without borrowing the answer.

  1. 01What Would We Predict Before Fitting Anything?Tomorrow's water demand is unknown. Yesterday's demand gives Botie a prediction he can actually make, and a reference that a more complicated forecast should improve on.
    Build with Botie →Read the lesson
  2. 02Building Inputs From Earlier DaysA prediction row needs information available before its target. We will turn one ordered record into rows of earlier readings without accidentally including the answer.
    Build with Botie →Read the lesson

Patterns That Repeat or Move

A series can repeat, drift, or react to recent observations. We will separate those behaviours before choosing a rule for updating a forecast.

  1. 03Separating Repetition From ChangeA demand increase might be a lasting change or a normal part of the weekly schedule. Comparing like positions in a cycle helps us distinguish those possibilities.
    Build with Botie →Read the lesson
  2. 04Updating a Forecast Without Starting OverBotie needs to respond to new demand without letting one unusual day replace everything he knows. A weighted update lets him control how quickly his estimate changes.
    Build with Botie →Read the lesson

Fitting and Testing Forecasts

A useful forecast must work with the information available at its origin. Fit a small model, then recreate that boundary on successive later days.

  1. 05Learning From a Series' Own PastA trailing average gives every lag the same influence. An autoregressive model fits different coefficients for the earlier readings and then uses that same rule to predict ahead.
    Build with Botie →Read the lesson
  2. 06Testing a Forecast as Time Moves ForwardOne convenient cutoff can hide a forecasting failure. We will replay several prediction dates, preserving what was known at each one, and compare methods on matching targets.
    Build with Botie →Read the lesson