tsforge — automatic, evidence-based forecasting for R

Give it a series. It validates, diagnoses, backtests a registry of models with rolling-origin validation, selects the empirical winner for this series and horizon under the configured validation procedure, forecasts with intervals, explains, and builds a Markdown/Quarto dashboard with interactive charts.

There is no universal best method — and this package will never claim one.

library(tsforge)
result <- auto_ts(data = sales, date = "date", value = "sales", horizon = 12)
summary(result)
plot(result)
ts_dashboard(result)   # static PNGs + plotly widgets

Why tsforge

  • Evidence, not preference. The winner is the backtest winner. Ties, failures and unavailable models stay visible for audit.
  • Honest metrics. 13 error metrics plus interval coverage; MAPE is NA whenever actuals contain zeros — never silently computed.
  • Extensible. register_model() adds your method without touching core.
  • Light base. Core runs on base R; ETS, GARCH, VAR, Prophet, ML and plotly are optional and capability-probed.

Status

Version 1.1.1: 29 registry models (+2 runtime ensembles), 63 exported functions, 15 runnable examples; quantiles, MinT, xreg, monitoring, streaming, Shiny, PMML, tidy; interactive dashboards; R CMD check clean. Quarto website (this site).