Time Series Forecasting Workflow

A practical flowchart for model design, diagnostics, and selection for time series forecasting problems.

Core Workflow

flowchart TD A[Define problem] --> D[Initial EDA] D --> E{Problem structure?} E -->|One target series| F[Univariate branch] E -->|Multiple target series| G[Multivariate branch] E -->|One target with exogenous inputs| H[Dynamic regression branch] F --> K[Fit benchmarks and candidates] G --> K H --> K K --> L[Residual and stability diagnostics] L --> M{Diagnostics adequate?} M -->|No| N[Respecify model] N --> K M -->|Yes| O[Time-series cross-validation] O --> P[Forecast evaluation] P --> Q{Best parsimonious model?} Q -->|No| K Q -->|Yes| S[Finalize forecast]

Diagnostics & Model Selection

flowchart TD B{Problem structure} %% Univariate B --> |Univariate| U1[Trend and stationarity] U1 --> U2[Seasonality checks] U2 --> U3[Variance diagnostics] U3 --> U4[Breaks and interventions] U4 --> U5[Special dependence checks] U5 --> Z[Candidate model list] %% Multivariate B -->|Multivariate| M0[System diagnostics] M0 --> M1{Integrated?} M1 -->|No| M2[VAR] M1 -->|Yes| M3[Cointegration tests] M3 --> M4{Cointegrated?} M4 -->|Yes| M5[VECM] M4 -->|No| M2[VAR in differences / differenced VAR] M2 --> Z M5 --> Z %% Dynamic regression B -->|Exogenous inputs| X0[Input-output diagnostics] X0 --> X3{Regression form} X3 -->|Lagged regressors| X4[ARDL / UECM / ECM] X3 -->|Regression with ARMA errors / ARIMAX| X5[ARIMAX / dynamic regression] X4 --> Z X5 --> Z

Model Family Selection

Condition Candidate models
Stationary linear AR / MA / ARMA
Trend / nonstationary ARIMA
Seasonal SARIMA / ETS
Multiple seasonality TBATS / BATS / MSTL
Exogenous inputs ARDL / ARIMAX
Multivariate VAR / VECM
Changing variance ARCH / GARCH / stochastic volatility / Generalized Autoregressive Score (GAS)
Long memory ARFIMA / FIGARCH
Nonlinear / regime TAR / SETAR / STAR / Markov switching
Nonlinear ML Neural network autoregression (NNAR)
Latent components / unobserved states State-space / structural time series / Kalman filter
Mixed frequency MIDAS
Intermittent demand Croston's method / TSB
Contact
LinkedIn | muratkoptur.com