Correlations and puzzles

Better still, the same things that moved gold prices in the 1980s and 1990s still move it now - the chart, right, shows there's little sign of these correlations breaking down.

You might think this solves the puzzle of explaining moves in the gold price. Not quite. The problem is that there's no nice theoretical justification for the correlations in our table. Facts and theory aren't consistent.

Before explaining this, here are some correlations with gold price moves that do make sense.

- Global share prices. A rise in dividend yields is associated with a fall in the gold price. The reason for this is simple. Higher yields on shares are usually a sign of higher expected returns. Because gold is an alternative to equities, expected returns on gold must therefore also rise, to induce the marginal investor to hold it. For expected returns to rise, the gold price must fall.

- Oil prices. Oil and gold move together. This could be because both benefit from 'geo-political uncertainty' or from higher inflation expectations. Or maybe it's because oil exporters spend some of their revenues on gold.

- Monetary growth. There are two measures of this that matter: bank deposits in the OECD region and world foreign exchange reserves, many of which are held by Asian central banks. If both rise by one standard deviation, gold prices rise an average of 9 per cent. This is a nice, old-fashioned monetarist story - when people have excess cash, they buy gold.

- The US dollar. A fall in the dollar is associated with a rise in gold. This could be because a weak dollar is a sign that investors are nervous about financial assets, and so want gold. Or maybe they anticipate that a weak dollar will raise Asian central banks' FX reserves and thus the demand for gold.

So much for the correlations that make sense. Now for some puzzling ones.

- Industrial production. The stronger this is in developed economies, the weaker is the gold price, controlling for other things. This might be because a weak economy is associated with uncertainty, which is good for gold. But why should this effect be powerful enough to offset what should be a positive income-elasticity of demand for gold?

- Global inflation. Higher inflation is bad for gold. That's not what you'd expect.

- The Fed funds rate. In theory, higher expected returns on US cash should mean higher expected returns on gold - and hence a lower gold price. But the opposite is true.

- US corporate bond yields. For the same reason, higher corporate bond yields should mean a lower gold price. But, again, the opposite is true. Perhaps this is because higher corporate yields are a sign of nervousness, which benefits gold.

The fact that there's no obviously satisfactory theoretical explanation for some correlates of the gold price means it's more than usually dangerous to use our table for forecasting.

For what it's worth, though, I reckon consensus-ish forecasts for the factors in the table point to gold dropping just below $500 by the year-end.

If you're bearish of the dollar or equities, though, or expect big global monetary growth, our table suggests you should be more optimistic.

However, even if the 1984-2005 relationships continue to hold, this forecast is still subject to a big margin of error, of $30. That's because gold is so volatile that a lot of its moves can't be related to macroeconomic factors.

The message here, I guess, is that we shouldn't trust neat, glib stories about the gold price. The facts are rather messier.

Estimated on annual changes, January 1984 to October 2005. *Except for *, which are percentage-point changes. All variables are statistically significant at the 1 per cent level, except for the Fed funds rate and Baa-rated bonds, which are significant at 5 per cent. The table ignores the intercept, of -9.18; this implies that, if all the variables in the table were zero, the gold price would fall by 9.18 per cent.