Do Investors Learn? Evidence from a Gold Market Anomaly. 1997
This 1997 paper is the first to look at whether gold conforms to the semi-strong form of the efficient markets hypothesis. It uses a portfolio of gold mining stocks as a possible source of information that the market may not be incorporating into the gold price instantaneously.
The one month lagged return of the gold portfolio is regressed on gold returns and based on OLS and GARCH estimates there is a statistically significant relationship. A 1% higher than average return for the portfolio of stocks predicts a .067% higher than average return for gold over the full sample under the GRACH model.
The authors then test whether this anomaly is persistent over the whole sample using monthly data. As this fact was first discussed in the Wall street journal in 1979 they test whether it dissipates after this time, testing whether the market learns. They find that pre-1979 the relationship is much stronger and statistically significant. After the anomaly is discussed the relationship becomes very small and insignificant at all levels. They also find that a structural break occurs at that point using a chow test.
Weekly data reinforces this finding, as the lagged return on the gold stock portfolio becomes insignificant in the second half of the sample but gold’s relationship to the concurrent portfolio return increases in significance dramatically. This implies that information is being incorporated more rapidly as the market learns and thereby becomes more efficient.
Using the knowledge of the anomaly they then test whether it is possible to trade gold based on it and beat the market. If the gold stock’s portfolio return is in the upper quartile in the previous month the trading rule buys gold – holding it until there is a sell signal. Gold is sold and replaced with a portfolio of S+P 500 stocks if the gold stocks portfolio return is in the bottom quartile. Over the full sample the trading rule is shown to outperform a buy and hold strategy even after deducting trading costs. In subsample analysis even after the announcement the trading rules still out performs a buy and hold strategy.
The paper concludes that over the period the gold market was not semi-strong form efficient but it did seem to be learning.
Data: Month-end gold price, PM Fixing. A portfolio of equally weighted gold producing stocks from the NYSE and ASE composed of 37 firms. S+P 500 for common stocks. August 1971 – End Dec 1992.
Methodology: OLS, GARCH models, chow tests
Citation: McQueen, Grant, and Steven Thorley. “Do investors learn? Evidence from a gold market anomaly.” Financial Review 32.3 (1997): 501-525.
Abstract: This study finds evidence that supports the investor learning hypothesis using data from the gold market. Consistent with conventional wisdom, the prior returns on an equally-weighted portfolio of gold-producing stocks are found to predict gold returns. However, the predictive power is shown to have diminished since the first public discussion of the anomaly, a finding consistent with the investor learning hypothesis.