ARDL Bounds Testing for Economic Time Series Modelling Training Course

Course overview

Economic and financial analysts frequently need to estimate relationships where variables are measured in a mixture of levels and first differences: GDP and inflation, credit growth and interest rates, exchange rates and trade balances, or earnings and macroeconomic drivers. Conventional cointegration procedures can be unsuitable when samples are short, integration orders differ, or the team needs both long-run estimates and short-run adjustment dynamics. This course equips participants to decide when an autoregressive distributed lag (ARDL) model and the Pesaran-Shin-Smith bounds test are defensible, rather than applying cointegration tests mechanically.

Participants build ARDL specifications from economic hypotheses, select lag orders, test for a long-run relationship using bounds-testing critical values, and estimate long-run coefficients and error-correction models. They learn to distinguish I(0), I(1), and prohibited I(2) series; interpret the unrestricted error-correction representation; conduct residual, stability, and functional-form diagnostics; and report results with appropriate cautions. Practical work covers EViews, Stata, R, and Gretl workflows, including reproducible model output, coefficient tables, diagnostic summaries, and charts.

The programme is delivered through instructor-led econometrics sessions, guided software labs, and a continuing applied case using macroeconomic and financial time series. Participants work through the full modelling sequence: data inspection, unit-root screening, ARDL estimation, bounds testing, diagnostic repair, interpretation, and management-ready reporting. Each participant leaves with a documented ARDL bounds-testing model, annotated code or command log, diagnostic evidence, and a concise technical briefing that can be adapted to a live workplace forecasting, policy, valuation, or risk-analysis assignment.

It is particularly suited to analysts who already work with time-series data and need a disciplined alternative to single-equation regression or standard cointegration approaches when sample size and integration properties require closer attention.

Course objectives

By the end of this course, participants will be able to:

• Classify economic time series as I(0), I(1), or potentially I(2) using plots and unit-root test evidence
• Specify an ARDL(p,q) model from an economic hypothesis and a defined dependent-variable relationship
• Select defensible lag orders using AIC, BIC, residual diagnostics, and economic timing assumptions
• Conduct the Pesaran-Shin-Smith bounds test and interpret lower-bound, upper-bound, and inconclusive results
• Estimate long-run coefficients and an error-correction model from an unrestricted ARDL specification
• Diagnose serial correlation, heteroskedasticity, non-normality, functional-form error, and parameter instability
• Compare ARDL results across EViews, Stata, R, and Gretl outputs using equivalent model specifications
• Produce a reproducible ARDL modelling report with tables, diagnostic charts, interpretation, and limitations

Benefits of attending

Target competencies

Who should attend

• Economic Analysts — who model macroeconomic indicators and need defensible short-run and long-run estimates
• Financial Analysts — who assess how rates, inflation, exchange rates, or output affect financial outcomes
• Central Bank and Policy Analysts — who evaluate transmission mechanisms and policy-variable relationships
• Credit Risk Analysts — who link portfolio performance or default indicators to macroeconomic drivers
• Treasury Analysts — who analyse interest-rate, liquidity, currency, and funding relationships over time
• Research Economists — who require a practical cointegration workflow for small and mixed-order samples

Requirements and prerequisites

Participants should be comfortable reading regression output and working with spreadsheet or statistical-software datasets. Familiarity with ordinary least squares, regression coefficients, hypothesis tests, p-values, confidence intervals, logarithms, and basic time-series terms such as lag and trend is assumed. Some prior exposure to stationarity or unit-root testing is helpful but not essential; these concepts are refreshed before ARDL modelling begins. Participants should bring a laptop capable of running EViews, Stata, R/RStudio, or Gretl. Advanced econometrics, matrix algebra, programming, vector error-correction models, and prior ARDL experience are not required.

Training methodology

The instructor introduces each modelling decision with a short economic rationale, then demonstrates the equivalent workflow in statistical software. Participants complete guided labs using a common macro-financial dataset, progressing from data transformation and unit-root screening to ARDL estimation, bounds testing, error-correction interpretation, and diagnostics. Small-group review sessions focus on choosing lags, resolving inconclusive bounds results, and challenging model assumptions. The final workshop requires each participant to assemble a reproducible model file and a decision-focused results briefing for a defined business or policy question.

Course outline

Day 1: ARDL foundations and time-series preparation

• Economic questions suited to autoregressive distributed lag models
• ARDL model structure, notation, and dynamic regression intuition
• Levels, first differences, logarithms, growth rates, and elasticity interpretation
• Visual inspection of trends, breaks, seasonality, and outliers
• Stationarity, integration order, and the I(2) exclusion rule
• Augmented Dickey-Fuller and Phillips-Perron unit-root testing
• Data import, date indexing, transformations, and missing-value checks in econometric software

Workshop: Participants prepare a macro-financial dataset, document transformations, and create an integration-order screening table for the case-study variables.

Day 2: ARDL specification and lag selection

• Selecting dependent and explanatory variables from an economic transmission hypothesis
• Restricted and unrestricted ARDL(p,q) specifications
• Intercepts, deterministic trends, seasonal controls, and exogenous dummy variables
• Maximum lag selection based on sample size and data frequency
• Akaike, Schwarz Bayesian, and Hannan-Quinn information criteria
• General-to-specific lag reduction and economic timing constraints
• Multicollinearity, over-parameterisation, and degrees-of-freedom risks

Workshop: Participants estimate candidate ARDL models, compare lag structures, and justify a preferred specification using a lag-selection worksheet.

Day 3: Bounds testing and long-run inference

• Unrestricted error-correction model representation of ARDL
• Pesaran-Shin-Smith bounds-test null and alternative hypotheses
• F-statistic calculation and joint significance of level variables
• Lower-bound, upper-bound, and inconclusive critical-value decisions
• Critical-value cases for intercepts and deterministic trends
• Small-sample considerations and bootstrap bounds-testing options
• Long-run coefficient estimation and long-run multiplier interpretation

Workshop: Participants run a bounds test on the case model, record the critical-value decision, and produce a long-run coefficient table with economic interpretation.

Day 4: Error correction, diagnostics, and model stability

• Deriving the error-correction term from the selected ARDL model
• Speed of adjustment, sign, magnitude, and statistical significance
• Short-run differenced coefficients and distributed-lag effects
• Breusch-Godfrey serial-correlation testing and residual correlograms
• Heteroskedasticity, normality, and Ramsey RESET specification tests
• CUSUM and CUSUM of squares parameter-stability tests
• Structural breaks, intervention dummies, and model re-specification

Workshop: Participants diagnose a deliberately flawed ARDL model, apply a justified re-specification, and prepare a before-and-after diagnostic summary.

Day 5: Applied reporting and workplace model deployment

• Interpreting ARDL output for forecasting, policy, risk, and valuation decisions
• Dynamic multipliers and adjustment-path charts
• In-sample fit, holdout evaluation, and forecast-error measures
• Robustness checks using alternative lag orders and variable definitions
• Comparing ARDL with Engle-Granger, Johansen, and VAR approaches
• Reproducible scripts, command logs, model documentation, and version control
• Technical reporting of assumptions, limitations, and bounds-test caveats

Workshop: Participants complete an end-to-end ARDL case report containing model selection evidence, bounds-test results, diagnostics, long-run findings, and an action-oriented management briefing.

Tools & standards covered

EViews, Stata, R, Gretl

A typical training day

Live online deliveries follow the same structure in the East Africa Time zone, with shorter screen blocks and longer breaks.

What the fee includes

How you can take this course

Certification

Participants who complete the full five days receive the Skillset Development Certificate of Completion, stating the course title, course code, dates and delivery format — suitable for professional-development records and employer reimbursement.

Frequently asked questions