In their February 2017 paper entitled “Bayesian Model Averaging, Ordinary Least Squares and the Price of Gold”, Dirk Baur and Brian Lucey analyze a large set of factors that potentially influence the price of gold via two methods: Ordinary Least Squares (OLS, scatter plot) and Bayesian Model Averaging (BMA, accounting for model uncertainty). They include as potential influencers three other precious metals futures, crude oil spot and futures, two commodity indexes, U.S. and world stock indexes, currency exchange rates, 10-year U.S. Treasury note (T-note) yield, U.S. Federal Funds Rate (FFR), a volatility index (VIX) and U.S. and world consumer price indexes. To test robustness of influencers, they consider: (1) subsamples to test consistency over time; (2) daily and monthly measurements to test consistency across sampling frequencies (except consumer price indexes, available only monthly); and, (3) contemporaneous and one period-lagged (predictive) relationships. Using daily and monthly prices for the specified assets during January 1980 through September 2016, they find that:
- OLS and BMA often yield similar results, but the latter helps identify inconsistencies in relationships between gold and its potential influencers.
- Only a few factors consistently influence contemporaneous and future gold prices.
- Precious metals, currencies and equity markets are important.
- T-note yield, FFR and consumer price indexes are not, with relationships varying considerably across the four decades in the sample period.
- Effects of potential influencers differ considerably between daily and monthly measurement frequencies.
- In general, the powers of the potential influencers to predict gold price behavior are poor based on a lag of one day or one month. Exceptions are oil prices for a one-day lag and the U.S. consumer price index for a one-month lag.
In summary, evidence from an array of consistency tests across a large number of variables offers little support for belief that the price of gold is usefully predictable at daily and monthly horizons.
Cautions regarding findings include:
- Testing many potential influencers on the same gold data introduces snooping bias, such that the performance of the best predictors overstates expectations.
- The study is academic and does not test any strategies for timing an allocation to gold based on findings.
