GARCH LAB

INTERACTIVE EXPERIMENT · GARCH(1,1)

Volatility after a shock

Change the size of a one-day return shock or the persistence of conditional variance. Compare the result with a fixed reference path.

Parameters

Multiples of daily volatility
α + β. The experiment fixes α at 0.08.

Expected variance path

Daily variance in percent squared. Navy shows your settings. Dashed gold shows the reference (3σ shock, persistence 0.96).

How it works

A 1% long-run daily volatility sets baseline variance to 1%². After a one-off shock of size z standard deviations, next-day variance is v₁ = 1 + α(z² − 1). From day one, the expected deviation from baseline decays at rate ρ = α + β per trading day. The half-life refers to that variance deviation.

E[vₕ] = 1 + α(z² − 1)ρh−1 · h ≥ 1

Methods notes

Practical checks for empirical finance and market-data research.

Volatility persistence

Compare two settings with the same shock size and different persistence. Read the half-life as the decay of excess expected variance. The shock size sets the initial displacement, while persistence governs its subsequent decay in this experiment.

Forecast accuracy and portfolio decisions

Evaluate forecast loss and portfolio performance on the same timestamped out-of-sample observations. Inspect turnover, weights, and transaction-cost assumptions before attributing a portfolio result to superior forecasting.

Distributional monitoring

Specify the reference window, evaluation window, and alert threshold before comparing alerts. Report false alarms, missed events, and response timing alongside any change in portfolio risk.

Cross-market event calendars

Record event timestamps and local time zones. Align overnight events and market holidays to each venue’s first eligible session before computing cross-market responses.

Read related papers at GARCH Hub ↗

Prepare a replication package

  1. Record the manuscript version and map each table and figure to its generating script.
  2. Document raw-data sources, extraction dates, transformations, and access terms. Provide permitted extracts or download instructions.
  3. Pin the software environment and give one execution order from inputs through final outputs.
  4. Re-run from a clean directory and compare key results with the manuscript.
  5. Archive a dated snapshot and record a SHA-256 digest for the archive.
README outline
Paper and manuscript version
Software and runtime
Data sources, access, and licenses
Directory map
Execution order
Tables and figures generated
Known limitations and expected runtime

See the World Bank DIME Analytics Data Handbook and AEA Data Editor package guidance.

Public data starting points

FRED

Macroeconomic and financial series. For vintage-sensitive work, record the release vintage and observation date.

BIS Data Portal

Banking, credit, and cross-border financial statistics. Record the reporting basis and units when you download.

SEC EDGAR

US company filings. Save accession identifiers and filing dates when working with point-in-time disclosures.