Time Series Econometric Forecasting Training Course

Participants can prepare traceable EViews workfiles for repeatable forecasting analysis.

• Importing Excel, CSV and database extracts into EViews workfiles
• Workfile frequency and dated-page settings for monthly and quarterly series
• Series transformations: logarithms, differences, growth rates and annualisation rules
• Missing-value codes, outliers and breaks documented before estimation
• Aligning release calendars and avoiding future-information contamination
• Building reproducible EViews programs, named objects and audit notes

Participants can diagnose data patterns and formulate initial model choices from visual evidence.

• Time plots, seasonal plots and correlograms used for initial diagnosis
• Distribution checks using histograms, quantiles, skewness and kurtosis
• Structural-break candidates identified through plots, event dates and residual patterns
• Trend-cycle separation with moving averages and Hodrick-Prescott filtering
• Seasonal decomposition by additive and multiplicative component structures
• Data-definition logs recording units, sources, transformations and revisions

Participants can estimate and interpret regression-based forecasts with trend and seasonal effects.

• Ordinary least squares forecasts with deterministic trend and seasonal dummies
• Intercept, slope and dummy coefficients interpreted in forecasting terms
• Creating forecast equations and dynamic versus static forecast paths
• Prediction intervals derived from coefficient and disturbance uncertainty
• Forecast tables exported from EViews equations and graph objects
• Benchmark comparisons against naive forecasts using common evaluation windows

Participants can test regression assumptions and decide whether model revision is required.

• Residual autocorrelation diagnosed with correlograms, Durbin-Watson and Breusch-Godfrey tests
• Conditional heteroskedasticity tested using ARCH LM procedures
• Normality checks using Jarque-Bera statistics and quantile plots
• Functional-form errors assessed with Ramsey RESET and residual plots
• Influential observations examined through hat values, Cook's distance and residuals
• Diagnostic findings translated into model revision decisions

Participants can test for unit roots and apply justified transformations before forecasting.

• Unit roots explained through stochastic trends and forecast consequences
• Augmented Dickey-Fuller tests specified with intercept, trend and lags
• KPSS stationarity tests interpreted alongside unit-root evidence
• Phillips-Perron tests compared where serial correlation remains uncertain
• First differences, log differences and seasonal differences selected appropriately
• Transformation decisions recorded with economic rationale and test results

Participants can identify, estimate and forecast autoregressive moving-average models in EViews.

• Autoregressive and moving-average processes represented with lag-operator notation
• ACF and PACF patterns used to propose ARMA orders
• Stationarity and invertibility conditions checked before model acceptance
• ARMA equations estimated in EViews using maximum likelihood
• Information criteria compared: AIC, BIC and Hannan-Quinn
• One-step and multi-step forecasts generated from fitted equations

Participants can specify seasonal forecasting models and account for calendar-driven variation.

• Seasonal ARIMA notation specified using nonseasonal and seasonal orders
• Seasonal unit roots identified through seasonal correlograms and tests
• Calendar regressors built for trading days, holidays and moving events
• X-13ARIMA-SEATS adjustment options reviewed for official-style series
• Seasonal forecast profiles checked against operational demand patterns
• Monthly and quarterly models compared using matched holdout periods

Participants can apply the Box-Jenkins process to select parsimonious ARIMA specifications.

• Box-Jenkins identification follows plots, correlograms, tests and parsimonious specifications
• Candidate ARIMA models estimated with consistent sample windows
• Residual white-noise tests determine whether serial structure remains
• Parameter significance assessed without relying solely on p-values
• Parsimony balanced against forecast accuracy and economic plausibility
• Model-selection records capture rejected alternatives and decision criteria

Participants can compare models using out-of-sample evidence and formal forecast accuracy tests.

• Rolling-origin evaluation reproduces repeated forecasts available at each decision date
• Fixed-origin holdouts distinguish development samples from final test samples
• Diebold-Mariano tests compare forecast accuracy between competing models
• Error metrics selected according to scale, zeros and decision costs
• Forecast bias assessed by horizon, season and business regime
• Champion-challenger scorecards rank models using pre-agreed criteria

Participants can estimate exponential smoothing models and compare them with ARIMA forecasts.

• Simple exponential smoothing estimated for level-only demand series
• Holt's linear trend method fitted with damped trend options
• Holt-Winters seasonal methods specified for additive and multiplicative patterns
• Smoothing parameters optimised using in-sample or holdout error
• ETS forecasts compared with ARIMA under identical evaluation designs
• Method choice aligned with data behaviour and maintenance capacity

Participants can model exceptional events and assess their effect on forecast reliability.

• Intervention variables model policy changes, strikes, promotions and one-off shocks
• Pulse, step and ramp dummies specified with clear event timing
• Outlier detection distinguishes additive outliers from level shifts
• Chow tests assess known breaks at defined dates
• Rolling coefficients reveal parameter instability over successive estimation windows
• Scenario assumptions separate temporary disruptions from persistent structural changes

Participants can model forecast drivers with distributed lags and appropriate timing controls.

• Distributed-lag regressions connect forecasts to economic and operational drivers
• Contemporaneous, lagged and leading variables screened for information timing
• Almon polynomial lags reduce parameter proliferation in long lag structures
• Autoregressive distributed lag models capture persistence and driver effects
• Multicollinearity diagnosed using correlation matrices and variance inflation factors
• Driver forecasts documented separately from the dependent-variable model

Participants can estimate long-run relationships and short-run adjustment models for nonstationary series.

• Cointegration distinguishes stable long-run relationships from spurious regressions
• Engle-Granger two-step testing estimates residual-based cointegration evidence
• Error-correction models combine short-run changes with equilibrium adjustment
• Johansen tests identify cointegration rank within multivariate systems
• Long-run coefficients interpreted with units, elasticities and restrictions
• Macroeconomic relationships tested for breaks before forecasting deployment

Participants can estimate VAR models and interpret multivariate forecast dynamics.

• Vector autoregressions model feedback among rates, inflation, output and credit
• Lag length selected with information criteria and residual diagnostics
• Granger-causality tests assess incremental predictive content, not structural causation
• Impulse-response functions trace shocks through jointly estimated variables
• Forecast error variance decomposition identifies principal forecast uncertainty sources
• VAR forecasts produced conditionally under stated paths for exogenous variables

Participants can estimate conditional volatility models for financial return series.

• ARCH effects identified from squared-residual correlograms and LM tests
• GARCH models estimate time-varying conditional variance for financial returns
• Mean equations paired with variance equations and distribution choices
• Asymmetric volatility modelled using EGARCH and threshold GARCH specifications
• Volatility forecasts converted into risk limits and interval assumptions
• Value-at-Risk inputs distinguished from full market-risk governance requirements

Participants can combine model outputs and document scenario or judgement-based forecast adjustments.

• Forecast combinations use equal weights, inverse-error weights and regression weights
• Ensemble candidates compared against individual-model benchmark performance
• Judgmental adjustments documented with source, rationale, magnitude and approval
• Scenario forecasts condition models on interest-rate, exchange-rate and volume paths
• Prediction intervals communicated separately from scenario ranges and stress cases
• Override governance prevents unsupported management adjustments to published forecasts

Participants can communicate forecast evidence clearly and establish an effective production process.

• Forecast packs present assumptions, accuracy, intervals and decisions required
• Fan charts show uncertainty widening across longer forecast horizons
• Actual-versus-forecast dashboards distinguish data revisions from model errors
• Forecast commentary explains driver movements without overstating statistical certainty
• Model inventory fields record owner, purpose, frequency and approval status
• Production calendars assign data cut-offs, reruns, reviews and release dates

Participants can establish controls that support reliable, reviewable and maintainable forecasting models.

• Model-risk controls cover versioning, access, review and change approval
• Reproducible EViews programs separate raw data, transformations and outputs
• Backtesting schedules define triggers for recalibration, replacement and retirement
• Independent review checks data lineage, assumptions, diagnostics and reporting
• Documentation meets internal audit needs for material planning models
• Forecast operating procedures assign accountable owners and escalation routes

Participants can build, review and present a complete forecasting package for a defined business or policy decision.

• Business question framed with target series, horizon, users and decisions
• Source data imported, documented and transformed within an EViews workfile
• Benchmark and candidate models estimated using justified specifications
• Rolling-origin backtests compare accuracy, bias and interval performance
• Diagnostics, assumptions and model limitations recorded for reviewer challenge
• Forecast package presents baseline, scenarios, uncertainty and implementation controls
• Peer review tests reproducibility, decision relevance and governance readiness