Experienced investors say that gold is ready to rally when nobody cares and that it's wise to sell when your hairdresser has a conversation about it with your taxi driver. It's not a secret that nowadays hairdressers and taxi drivers find their ideas on Google (NASDAQ:GOOG).
Here are the Google Trends charts for gold price since 2005:
The first chart is the search volume: it is an indicator of the number of requests from people who want to read something about gold price. We can infer with little risk that it is in direct relation with the number of investors thinking of buying or selling a metal related asset.
The second chart is the Google news volume: it is an indicator of the number of Google-registered publications about gold price. Writing financial information and publishing it on ranked websites takes time and/or money. So we can infer with little risk that it is in relation with the gains that opinion leaders are expecting from the next move in gold price.
Here is the price of the Gold ETF GLD on the same period.
At first sight we are tempted to see a correlation between search volume peaks and GLD highs. The question is: do the peaks occur before or after the highs ?
Let's focus on 2011 and let's compare the search volume with the price.
There is a baseline around which search volume oscillates before jumping. This baseline is what information theory specialists name "noise." Let's look at some numbers if we normalize the values of search volume fixing the annual average at 1.
The average of the 6 first months of 2011 is 0.71.
The search volume on the week of august 7th is 2.16.
As gold trend was broken on the 23th we are tempted to affirm that a good predictive signal would have been sent when the search volume went above 3x its previous baseline.
There was a second higher peak on the week of September 25th at 2.18. This time a fall in gold price began on the 22th and it would probably have been no use as a predictive signal. In fact we can't be so sure because Google Trends doesn't give daily data older than 30 days.
Now let's have a look at the search volume for silver price on 2011, compared with silver ETF SLV.
The normalized average in search volume of the 3 first months of 2011 is 0.78.
The search volume on the week of April 24th is 2.04.
As silver trend was broken on may 2nd, the jump above 2.5x the "noise" level would have been a good signal.
There was a lower peak of search volume on the week of September 25th at 1.76. Like for gold, it would probably have been no use as a predictive signal this time.
We can wonder what we could see on a shorter time frame. The daily data for Google Trends are available for the last 30 days only. Moreover there is an natural weekly cycle in Google activity that masks weaker signals.
So nothing appears clearly on a daily basis for may 2012.
Of course, this short article is far from sufficient to draw a probabilistic conclusion and claim that a 2 or 3 ratio of search volume above the noise level is the right thing to watch. It has worked twice (for the biggest and first fall of each metal in 2011) and has given a "just too late" signal twice (for secondary falls). The aim here is neither an academic publication nor finding an investment strategy. However we can't deny a correlation between high search volume and gold and silver price. A steep rise in gold price or silver price search volume should be considered if not as a signal, at least as a warning.
Disclosure: I had a Google Advertising Professional certification in 2007 but didn't renew it and have always been independent of Google Inc. I own shares of physical precious metal funds: ZGLD, ZSIL, CEF, GTU. I am long CEF, GTU.
Bibliography: "Predicting the Present with Google Trends," Hyunyoung Choi & al, copyright Google Inc. 2009