Time Series Analysis Using R Training Course

Participants can prepare clean, consistently indexed data for forecasting models.

• Identifying duplicate timestamps, missing periods, and conflicting source records
• Treating missing observations through interpolation, imputation, and documented assumptions
• Aggregating transaction data into daily, weekly, monthly, and quarterly series
• Creating calendar variables for weekdays, holidays, month-end, and trading days
• Detecting extreme values using time plots, boxplots, and residual screening
• Documenting source lineage, transformations, and forecast-ready data definitions

Participants can identify the main temporal features that determine model choice.

• Distinguishing level, trend, seasonality, cycles, and irregular variation
• Seasonal subseries plots reveal within-year patterns and changing seasonal strength
• Lag plots expose serial dependence and nonlinear temporal relationships
• Heatmaps display demand concentration by weekday, week, or month
• Comparing related series through indexed values and aligned time plots
• Framing forecast questions with horizon, frequency, decision, and error costs

Participants can decompose time series and interpret trend and seasonal components.

• Classical additive and multiplicative decomposition for stable seasonal patterns
• STL decomposition using LOESS smoothing and adjustable seasonal windows
• Interpreting seasonally adjusted series for policy and management reporting
• Seasonal strength measures guide model specification and data transformation
• Comparing additive versus multiplicative seasonality through residual behaviour
• Extracting trend and seasonal components from fable decomposition models

Participants can create benchmark forecasts and evaluate forecast error appropriately.

• Mean, naïve, seasonal naïve, and drift forecasting benchmarks
• Training, test, and validation periods chosen without future-data leakage
• MAE, RMSE, MAPE, MASE, and RMSSE accuracy measures
• Selecting accuracy metrics according to scale, zeros, and decision costs
• Residual forecast errors compared across horizons and seasonal periods
• Benchmark results establish minimum performance standards for candidate models

Participants can use autocorrelation evidence to assess series dependence and stationarity.

• Autocorrelation function plots identify persistence at successive time lags
• Partial autocorrelation supports autoregressive order identification
• White-noise residuals and their implications for forecast adequacy
• Stationarity assessed through level behaviour, differencing, and unit-root tests
• Applying first differences and seasonal differences to nonstationary series
• Interpreting ADF and KPSS tests alongside graphical diagnostic evidence

Participants can fit and interpret exponential smoothing models for different demand patterns.

• Simple exponential smoothing for level-only short-horizon forecasting
• Holt linear trend method with damped trend alternatives
• Holt-Winters additive seasonality for stable seasonal variation
• Multiplicative error, trend, and seasonality ETS model components
• Automatic ETS selection using information criteria and residual diagnostics
• Comparing ETS forecasts against seasonal naïve benchmark performance

Participants can specify and fit non-seasonal ARIMA models in R.

• AR, MA, and ARMA processes represented through lag operators
• ARIMA p, d, q notation and differencing requirements
• Reading ACF and PACF patterns for tentative order selection
• Fitting ARIMA models with fable and forecast package functions
• AICc and BIC comparisons across plausible model specifications
• Interpreting autoregressive and moving-average coefficients without overclaiming causality

Participants can build seasonal ARIMA models for recurring calendar patterns.

• Seasonal ARIMA P, D, Q notation at defined seasonal frequencies
• Seasonal differencing removes repeating annual, monthly, or weekly patterns
• Seasonal ACF spikes identify candidate moving-average seasonal terms
• Fitting SARIMA models for monthly sales and weekly service volumes
• Checking overdifferencing through ACF behaviour and forecast instability
• Comparing seasonal ARIMA models with ETS and seasonal naïve forecasts

Participants can assess forecast models using residual tests and selection evidence.

• Residual time plots reveal unmodelled patterns, breaks, and variance changes
• Ljung-Box tests assess remaining autocorrelation in forecast residuals
• Histogram and Q-Q plots assess distributional assumptions for intervals
• Checking residual mean, variance, independence, and extreme observations
• Selecting parsimonious models that meet accuracy and diagnostic requirements
• Recording model decisions in a transparent forecast selection log

Participants can communicate forecast uncertainty and develop planning scenarios.

• Point forecasts, prediction intervals, and confidence intervals distinguished
• Producing 80 percent and 95 percent forecast intervals
• Interval width interpreted against planning tolerances and service targets
• Scenario forecasts based on alternative assumptions about future drivers
• Fan charts communicate uncertainty across multiple forecast horizons
• Escalation thresholds defined when realised values breach forecast intervals

Participants can model external predictors while accounting for time-series errors.

• Dynamic regression combines explanatory variables with autocorrelated error processes
• Creating lagged predictors for promotions, prices, weather, and policy changes
• Distributed lag models capture delayed effects of external drivers
• Testing predictor availability at the required forecast horizon
• Regression with ARIMA errors fitted using ARIMAX specifications
• Separating explanatory relationships from operational forecasting requirements

Participants can account for exceptional events and changes in time-series behaviour.

• Pulse interventions represent one-off disruptions, closures, and system failures
• Step interventions model permanent policy, tariff, or process changes
• Ramp functions capture gradual implementation and phased demand shifts
• Change-point analysis identifies structural breaks in level or trend
• Outlier treatment preserves auditability and avoids deleting material events
• Post-intervention forecasts distinguish temporary shocks from persistent changes

Participants can address forecasting problems with more than one seasonal pattern.

• Daily data with weekly and annual seasonal patterns
• Hourly demand series with intraday, weekly, and holiday effects
• Fourier terms represent long seasonal cycles within regression models
• STL-based forecasting handles complex recurring seasonal structures
• Aggregation choices balance forecast usefulness against high-frequency noise
• Comparing daily operational forecasts with monthly planning forecast requirements

Participants can produce coherent forecasts across organisational reporting hierarchies.

• Hierarchical series structures for products, regions, branches, and channels
• Bottom-up forecasting aggregates detailed forecasts into organisational totals
• Top-down allocation distributes aggregate forecasts using historical proportions
• MinT reconciliation produces coherent forecasts across all hierarchy levels
• Grouped time series manage overlapping product and geographic classifications
• Reconciling forecasts supports consistent budget and operational planning reports

Participants can establish ongoing forecast validation and model monitoring procedures.

• Rolling-origin cross-validation simulates repeated real forecast decisions
• Expanding and sliding windows test model stability over time
• Forecast bias monitored through mean error and tracking signals
• Accuracy dashboards compare models, horizons, segments, and forecast vintages
• Refit schedules account for new data and changing seasonal behaviour
• Model retirement criteria address persistent degradation and structural change

Participants can package forecasting results into usable reports for decision-makers.

• Quarto reporting combines code, plots, assumptions, and forecast tables
• Forecast narratives distinguish observed facts, model outputs, and judgement
• Decision tables translate forecast ranges into staffing and inventory actions
• Version control principles support reviewable forecasting scripts and assumptions
• Model cards record data scope, limitations, accuracy, and ownership
• Presenting forecast uncertainty without misleading precision or technical jargon

Participants can build, review and present a complete R-based forecasting analysis for a defined business problem.

• Define a decision-focused forecast question, horizon, metric, and stakeholder
• Prepare indexed data and document transformations, exclusions, and assumptions
• Diagnose temporal patterns using plots, decomposition, and autocorrelation evidence
• Fit benchmark, ETS, ARIMA, and regression-based candidate forecasting models
• Select a model through diagnostics, rolling-origin accuracy, and practical constraints
• Generate forecast intervals, scenarios, management visuals, and recommended actions
• Review peer analyses against a structured forecasting quality checklist