risk management

Predicting Gold using Currencies




The price of gold is a unique and interesting financial series, but not an easy one to predict. It is very volatile – with large, unexpected movements.

In this post we analyse its relationship with certain currencies to see whether they can provide an indication of the direction of gold movements.

There are various hypotheses surrounding relationships between gold and currencies. We are going to compare two of them:

  1. Gold behaves like commodity currencies.
  2. Gold behaves like safe-haven currencies.

Firstly, many believe that gold moves with the ‘commodity’ currencies. These are countries that possess large quantities of commodities or other natural resources, and their economies are highly dependent on the exports of these materials. We focus on Canada, Australia and New Zealand, which also happen to be some of the largest producers of gold itself.

On the other hand, others argue that in times of uncertainty or economic crisis, gold is considered a stable investment and is a popular safe-haven for wealth. Gold is free from interventions; is not controlled by a central bank and can’t be manipulated by interest rate policies. The most well-known ‘safe-haven’ currencies are Switzerland and Japan. Also, historically the Swiss Franc has been backed by gold reserves equating to 20% of its assets, hence emphasising the possible connection between the two.

We contrast these ideas by studying the relationship between each currency crossed with the US dollar (AUDUSD, NZDUSD, CADUSD, CHFUSD, JPYUSD) and the gold price (XAUUSD).

Evolution of the Series

Inspecting the price movements of the series from 2000, gold does appear to behave very much like the crosses during certain periods. Do these relationships really exist, or are our eyes deceiving us?

Price movements of the series from 2000

How strong are the relationships?

To answer this question we calculate the total series correlation using various return frequencies: from 1 day to 3 months. The correlations are positive without exception. In total the Swiss Franc and the Australian Dollar seem to be slightly more correlated than the rest.

In the 3-year rolling correlation of daily returns, we appreciate that the relationships are quite weak and not very constant. Also, in recent years the correlation has increased considerably in all crosses.

Simultaneous Total Correlation  Rolling Correlation with Gold 3 years

We put these values into perspective by implementing a trading strategy for gold based on the type of markets in the currencies. We take a long position in gold when the currency is bullish, a short position when the currency is bearish and do nothing when the currency market is lateral moving.

Gold Equities using Currencies splitnoiser Gold Rolling Equities using Currencies splitnoiser

The results are surprisingly good, reaching over 20% annual return on average. This suggests that the gold market type does coincide with the currency spot movements most of the time. Unfortunately, this is not a realistic strategy since the algorithm used to define the currency market types is forward-looking. That is, it uses future information to decide the direction of the spot movement.

But we can still use the results to compare the five crosses. The two safe-haven currencies achieve better overall results. The commodity crosses have some good moments but are inferior on average. This could indicate that gold is more related to Switzerland and Japan than the others, and should therefore be considered a safe-haven.

Are the relationships exploitable?

Correlation is a popular statistic, but not particularly useful in prediction since we have calculated it over the ‘simultaneous’ returns. In other words, we are comparing movements that occured at the same time. In order to predict gold, we need there to be a cause-consequence relationship and then be able to exploit it.

By shifting the gold series forward and comparing past currency returns with ‘future’ gold ones, we are able to analyse if the currencies have an impact on gold and can therefore be used to predict the series.


Since correlation is not affected by the movement’s magnitude, we can easily compare different return frequencies in the series. The important thing is to make sure the gold returns starts after the currency returns have concluded.

The correlations of these shifted series are hugely reduced compared to the simultaneous statistics. In some cases the relationships are inverted (negative correlation), implying that gold moves in the opposite direction and not with the currency as we would expect. It is worth noting that Japan shows a stronger cause-consequence relationship than the rest, especially for longer frequencies.

Shifted Correlations pcolor

Trading Strategies

Despite these poor impact relationships, we can implement two strategies to directly evaluate use of currencies to predict the direction gold is moving. Both strategies only use past information and are realistic in terms of their implementation.

Strategy 1: We trade in the direction the currencies have moved in the previous period. We have a series of equities for each pair of frequencies; each length of the past currency returns with a certain trade duration in the gold. Each day our position (quantity long or short) is the average of all the open trade estimates.

Strategy 2: We implement a generic, non-optimised trend-following strategy on the five currency crosses. We trade the gold according to the strength of the long or short position estimated for each individual currency.

In the first strategy we have the following 36 equity series using the movements in each currency cross. In general, the longer frequencies (of the past currency returns) are superior to the short term equities.

XAUUSD All Mov. Strategy Equities Grouped by Cause_legend

We summarise the results of strategy 1 for each currency, taking an average over all the equities. Surprisingly, despite the difference in the two methods used in the strategies, their results are quite similar.

Strategy 1:

Combined Frequencies Strategy Equities

Strategy 2:

God equities using currencies eclectic

These results are by no means spectacular (less than 2% average annual returns), but considering the strategy’s simplicity, the outcome is fairly promising. Grouping together the two different types of crosses, we can see that before 2009 the commodity crosses were slightly better at ‘prediciting’ gold. In more recent years, the safe-haven currencies have been superior in their predictions of the gold direction. This suggests that gold used to behave like a commodity, but since the financial crisis it has adopted a more safe-haven role.

Currency Set Strategy Equities all

We’ve seen that these crosses do have a close relationship to gold but they are not very constant or stable. This makes it difficult to exploit in a trading strategy. It’s clear that the possible success of using the relationships depends greatly on the accuracy of the trend detection in the underlying currency pairs. Since this is not an easy task, the possibilities of using currencies to predict gold are quite limited. However, the idea could be successful if used correctly; possibly as part of a more sophisticated strategy.

add a comment

awesome blog, great post. would love to hear your take along similar foundations (relationships between gold and comdollars/risk currencies), but perhaps in a manner that somehow trades the mean-reversion of the correlations? (if that makes sense!)

I have study strong in silver,that look detect around 15 sign by year.Next in silver is 20 march.If 19 march is up day or are in top should 20 march are sell day or inverse.
Last sign was 21 april.
Gold And oil emulated this sign but not Always,this is best silver.
20 march nex point.

I like your analysis and charts – can you explain which software you use to run those calculations?

B, thank you for your comment.

All of the analysis have been done using Matlab and almost entirely using personally programmed functions. The graphs were also created using Matlab’s plotting tools.

I found your post very interesting. I wonder if your results can improve using the 5 currencies simultaneously. Did you try any tests like this?

Thank you for your interest, xristica.

In this post I was interested to see which type of crosses had more predictive power for gold historically and so the tests were mostly on the individual crosses. However, I do agree it could be a good idea to combine factors and use various currencies at the same time. I imagine the more (relevant) factors involved, the better the outcome.

[…] Predicting Gold using Currencies [Quant Dare] The price of gold is a unique and interesting financial series but not an easy one to predict. It is very volatile with large sharp unexpected movements. In this post we analyse its relationship with certain cu… […]