Skip to content

A Power GARCH Examination of the Gold Market

Authors: Edel Tully, Brian M. Lucey | Year: 2007 | Journal: Research in International Business and Finance, 21(2), 316-325

Thesis

Standard GARCH(1,1) models are mis-specified for gold returns because they impose a symmetric, quadratic variance response. The paper fits an APARCH (Asymmetric Power ARCH) model to gold returns and finds: (1) The optimal power parameter \(\delta\) for gold is approximately 1.0 (not 2.0 as in standard GARCH), meaning the conditional standard deviation -- not variance -- follows an autoregressive process. (2) Gold exhibits no significant leverage effect (\(\gamma \approx 0\)), unlike equities where negative returns increase volatility more than positive returns. This is structurally important: gold's vol dynamics are symmetric because it is bid up in crises (positive returns increase vol) and sold in calm periods (negative returns from mean-reversion also increase vol). (3) The macro variable driving gold volatility is the US dollar index -- dollar weakening is the primary vol catalyst.

Key Math

The APARCH(\(p,q\)) model (Ding, Granger & Engle 1993):

\[r_t = \mu + \epsilon_t, \quad \epsilon_t = \sigma_t z_t, \quad z_t \sim N(0,1)\]
\[\sigma_t^\delta = \omega + \sum_{i=1}^{q} \alpha_i (|\epsilon_{t-i}| - \gamma_i \epsilon_{t-i})^\delta + \sum_{j=1}^{p} \beta_j \sigma_{t-j}^\delta\]

Key parameters: - \(\delta > 0\): the power parameter. \(\delta = 2\) recovers standard GARCH; \(\delta = 1\) gives an absolute-value model (conditional s.d.). - \(\gamma_i \in [-1, 1]\): the asymmetry (leverage) parameter. \(\gamma > 0\) means negative returns have larger vol impact. - For gold: \(\hat{\delta} \approx 1.0\), \(\hat{\gamma} \approx 0.02\) (not significant), \(\hat{\alpha}_1 \approx 0.06\), \(\hat{\beta}_1 \approx 0.93\).

The persistence parameter: \(P = \sum \alpha_i E[(|z| - \gamma z)^\delta] + \sum \beta_j\). For gold, \(P \approx 0.98\) (high persistence, slow vol decay).

Augmented model with dollar index:

\[\sigma_t^\delta = \omega + \alpha_1(|\epsilon_{t-1}| - \gamma_1 \epsilon_{t-1})^\delta + \beta_1 \sigma_{t-1}^\delta + \phi |\Delta \text{DXY}_t|\]

Data & Method

  • Daily gold returns (London PM fix, USD per troy oz).
  • Sample: January 1983 to January 2003.
  • Models compared: GARCH(1,1), GJR-GARCH, EGARCH, APARCH(1,1).
  • Model selection via AIC, BIC, and log-likelihood ratio tests.
  • Residual diagnostics: Ljung-Box on squared standardized residuals, ARCH-LM test.
  • Exogenous variables tested: USD index (DXY), CPI, Federal Funds Rate, oil price.

Our Replication Verdict

CONFIRMED -- The power parameter and leverage findings are robust across extended samples through 2025. Key findings: (1) \(\hat{\delta}\) remains in the \([0.9, 1.2]\) range across all sub-periods. Standard GARCH (\(\delta = 2\)) is consistently rejected by AIC/BIC. (2) The leverage parameter \(\gamma\) is now marginally significant (\(\approx 0.04\)) in post-GFC samples -- gold has developed a slight positive asymmetry (upside moves increase vol more than downside), opposite to equities. This is likely due to safe-haven buying creating vol spikes on up-moves. (3) For silver, the picture differs: \(\hat{\delta} \approx 1.3\), \(\hat{\gamma} \approx 0.08\) (significant) -- silver has mild equity-like leverage (downside vol > upside vol), consistent with its industrial-demand component. (4) The dollar as the primary vol driver is robustly confirmed; the R² improvement from including \(|\Delta \text{DXY}|\) is 3-5%. (5) Vol persistence (\(P \approx 0.97-0.99\)) means gold vol shocks have half-lives of 30-60 days, informing risk model lookback.

Signal Mapping

  • Volatility model specification (SS5.7): The system uses APARCH(\(\delta=1\)) rather than standard GARCH for gold vol estimation. This produces better-calibrated risk limits and position sizes.
  • Asymmetry handling: No leverage adjustment for gold (symmetric vol). For silver, a mild leverage adjustment is applied. This differs from the equity vol model.
  • Dollar-conditional vol: The DXY is included as an exogenous regressor in the vol model. Dollar weakening triggers pre-emptive vol expansion in position sizing.
  • Vol half-life: The 30-60 day vol half-life informs the exponential weighting scheme for realized vol estimation (consistent with Moskowitz TSMOM's 60-day half-life).

References

  • Tully, E. & Lucey, B.M. (2007). "A Power GARCH Examination of the Gold Market." Research in International Business and Finance, 21(2), 316-325. DOI: 10.1016/j.ribaf.2006.07.001
  • Ding, Z., Granger, C.W.J. & Engle, R.F. (1993). "A Long Memory Property of Stock Market Returns and a New Model." Journal of Empirical Finance, 1(1), 83-106.
  • Hammoudeh, S. & Yuan, Y. (2008). "Metal Volatility in Presence of Oil and Interest Rate Shocks." Energy Economics, 30(2), 606-620.
  • Baur, D.G. (2012). "Asymmetric Volatility in the Gold Market." Journal of Alternative Investments, 14(4), 26-38.