Why a Gold Factor Model Can Have Modest Explanatory Power

Summary

The discussion addresses a regression intended to explain gold using the VIX, the EUR/USD exchange rate, and a US real-rate index. The model uses ridge regression because its author selected a small, theory-informed set of factors, yet reports an adjusted R-squared of 36% and asks whether that is unexpectedly low. The reply assumes the regression uses contemporaneous returns for gold and the factors, making the exercise one of explaining gold-return variation rather than matching similar-looking price histories.

The response considers that explanatory power reasonable for a commodity affected by many influences beyond the selected variables. It gives seasonal Indian jewelry demand, volatility in developing-market currencies, and hyperinflation in major economies as examples of omitted drivers. The main lesson is that correlated trends in price-level charts do not imply that a few factors will explain most return variation. This is a brief interpretive answer, not a model comparison or empirical validation: it does not test the chosen factors, sample period, regression specification, or predictive performance. Its assessment depends on the stated assumption about contemporaneous returns.

Key ideas

• A regression on contemporaneous returns explains variation in changes, not similarity between price-level charts.
• A small set of plausible factors may leave substantial gold-return variation unexplained.
• Gold can respond to seasonal demand and currency or inflation conditions beyond the selected factors.
• The reported adjusted R-squared is interpreted as reasonable, but the reply does not independently validate the model.
• Explanatory fit alone does not establish predictive performance.

Tags

# Factor model for Gold has low adjusted R2


# Factor model for Gold has low adjusted R2












I am trying to build up a factor model for gold. To be able to identify the correct factors, I did a correlation analysis between a few factors vs gold and I integrated this analysis with what I saw in literature. I ended up with a set of factors as: VIX Index, EURUSD currency, H15X10YR Index (which is the US 10YR real rate). However, after running my ridge regression (I choose ridge due to the fact that I had already an idea of the factors and they are only 3, otherwise in case of many factor and no idea I would opt for lasso), my adjusted R2 is "only" 36%..am I missing something? after checking the time history I see that the gold has quite a similar trend with my factors, especially with the H15X10YR Index, but my level of "explicability" is only 36%. Can anybody highlight me on the reason of this please? Luigi

## Answer by kurtosis (score 2)

https://quant.stackexchange.com/a/57327

I will assume you are using factors and gold returns that are contemporaneous. With that setup, you are essentially trying to explain or decompose gold returns.

For an explanatory regression of a commodity (which is internationally traded), an $R^2$ of 36% is pretty good. Lots of factors can affect gold returns: Indian wedding season (a major effect on gold markets which you should not overlook), volatility in developing market currencies, and hyperinflation in large economies. You are omitting all of those effects. To still explain 36% of the variance is something to be proud of.

You may have expected more, but working with returns reveals how explaining changes in an asset isn't as simple as just showing a few similar-looking plots and exclaiming voila! I would not be upset at getting a 36% $R^2$ for my first model and one using only three factors.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.