Time Series Forecasting

Benchmarks, backtests and honest error tracking first, then a monthly forecasting rhythm, core models, ranges, reconciliation and specialist methods for demand, revenue and traffic forecasts.

Free · Mac (Apple silicon) and Windows · v1.0.5 · Other platforms

Then open the Marketplace in the app, search for Time Series Forecasting, and click Install. It goes into your own files, and works offline from then on.

What it installs.

1

Pillar

1

Area

50

Projects

200

Starter tasks

📊 Data & Analytics

Time Series Forecasting

Forecasting values like demand, revenue or traffic over time using seasonal and statistical models, and tracking forecast error honestly.

  1. Forecast purpose and decision brief
  2. Clean history of the target series
  3. Decomposing the series into trend and seasonality
  4. Naive and seasonal naive benchmark forecasts
  5. Forecast error metric selection
  6. Rolling-origin backtest harness
  7. Calendar and events table for forecasting
  8. Outliers and structural breaks log
  9. Forecast level and hierarchy decision
  10. Monthly forecast refresh cycle
  11. Forecast accuracy tracking log
  12. Forecast snapshot archive
  13. Forecast bias and tracking signal review
  14. Weekly demand forecast run for operations
  15. Forecast assumptions register
  16. Judgemental override log
  17. Quarterly forecasting model review
  18. Annual forecasting calendar and reforecast points
  19. Exponential smoothing models in practice
  20. ARIMA and seasonal ARIMA fundamentals
  21. Prediction intervals and probabilistic forecasts
  22. Forecasting with external drivers
  23. Prophet and decomposable models, tested honestly
  24. Gradient boosting with lag features
  25. Forecasting textbook study plan
  26. Intermittent demand forecasting methods
  27. Forecasting tool and library selection
  28. Forecast value added study
  29. Hierarchical forecast reconciliation
  30. Forecast combination trial
  31. Ranges instead of single numbers in forecast reports
  32. Automating the forecast run end to end
  33. New product forecasting by analogy
  34. Budget season revenue forecast delivery
  35. Peak season forecast preparation
  36. Post-season forecast retrospective
  37. First consensus forecast meeting
  38. Leadership forecast briefing with scenarios
  39. Inheriting someone else's forecast
  40. Shared forecasting process for a small team
  41. Forecasting study with a pre-registered evaluation
  42. Spreadsheet forecast for a solo analyst
  43. Website traffic forecast for a content team
  44. Contact centre volume forecast by interval
  45. Prediction interval calibration audit
  46. Benchmarking deep learning and foundation models
  47. Multiple seasonality in high-frequency data
  48. Forecasting through regime change
  49. Unconstrained demand from censored sales
  50. Forecasting thousands of series at scale

Where it starts.

Each project arrives with its purpose, milestones and first tasks written out. Here is how the first three read.

Forecast purpose and decision brief

Before any model is fitted, someone has to say which decision the forecast feeds: a stock order placed eight weeks ahead, a hiring plan, a quarterly revenue guide. That decision fixes the horizon, the time bucket, the lead time and the cost of being wrong in each direction, and most forecasting disputes trace back to these never being written down. A one-page brief settles them before effort goes into methods.

First tasks

  • Write down the decision the forecast will feed and who makes it
  • Ask the decision owner how far ahead and at what level they need numbers
  • Estimate what an over-forecast and an under-forecast each cost
  • Circulate the one-page brief and record the user's sign-off

Clean history of the target series

Models learn whatever the history says, including the month the definition of revenue changed and the week the warehouse system double-counted orders. Pulling at least two full seasonal cycles from one source, under a single consistent definition, is the groundwork every later project depends on.

First tasks

  • Write a one-sentence definition of the series and its source
  • Extract the full available history at the agreed time bucket
  • List gaps, duplicates and definition changes with their dates
  • Reconcile the cleaned yearly totals to a finance or system report

Decomposing the series into trend and seasonality

Looking hard at the data before modelling saves weeks. A time plot, a seasonal plot and an STL decomposition show whether the trend is steady or bending, whether the seasonal swing grows with the level, and how noisy the remainder is, which together point to the families of model worth trying.

First tasks

  • Plot the full history and mark known events on it
  • Draw a seasonal plot comparing each year on the same axis
  • Run an STL decomposition and save the three panels
  • Write a half-page note on what the decomposition shows

Make it yours.

Install Time Series Forecasting, then tell Aurum the date you're aiming for. It checks whether the plan fits the weeks you actually have.

Free · Mac (Apple silicon) and Windows · v1.0.5 · Other platforms

Blueprints are general information to help you organise your own work. They are not medical, financial, legal or other professional advice. See the Terms.