Lab Methodologies
These pages document the methods exposed in Macro by Mark Labs. They are written for readers who need to know what the tool is doing, what assumptions it makes, and where the current implementation stops.
The docs are part of the product contract. They should not describe planned ingestion, live monitoring, or model-native simulation as shipped work unless the runtime path already supports it.
Empirical Forecasting
ARIMA, VAR-family, Bayesian VAR, VECM, structural VAR, density, and factor short-horizon workflows.
Forecast Combination
Point, density, fan-chart, tournament, monitoring-snapshot, and scenario-conditional combination.
Stationarity Testing
ADF and KPSS tests with deterministic specification and lag choices made explicit.
Diagnostic Tests
ACF, PACF, Ljung-Box, Jarque-Bera, and Engle ARCH-LM checks for residual behavior.
SVAR Identification
Recursive, sign-restricted, long-run, proxy, and narrative sign-restricted structural analysis.
BVAR Posterior
Conjugate Bayesian VAR with Minnesota-style normal-inverse-Wishart shrinkage.
VECM & Cointegration
Johansen rank tests, maximum-likelihood VECM estimates, and rank-restricted forecasts.
Density Forecasting
Bootstrap, posterior, state-space, and explicitly labeled approximate Gaussian density paths.
Forecast Combination
Bates-Granger, equal weights, stacking, and Granger-Ramanathan point-combination methods.
Fan Charts
Central probability bands built from combined density forecasts and optional judgment adjustments.
Monitoring Snapshots
Rolling error metrics and threshold alerts computed from completed forecast runs.
Tournament Comparison
Aligned holdout comparisons with scorecards, Diebold-Mariano, Clark-West, and model confidence sets.
Public Status
The empirical and combination labs are no longer treated as blank methodology shells. The public docs now cover the main shipped paths. Some data-ingestion paths are still bounded: external forecast uploads, scheduled monitoring refresh, and live feed transport are not described as complete until the runtime supports them.