If Google has done nothing else, they have provided “abundance” to low budget data wranglers. Following the highly publicized (yet essentially trivial) Google Flu Trends, who have generated a neverending series of studies to show “Google Searches Predicts X”, and similar discoveries.
This week we have another one, “The Predictor Impact of Web Search Media On Bitcoin Trading Volumes” by Martina Matta, laria Lunesu, and Michele Marchesi.
The investigators used data from Google trends and public data on Bitcoin transactions to look at possible correlations and predictions during 2014 and 2015. Specifically, they compared the frequency of searching for the term “bitcoin” to several measures of global Bitcoin activity.
The results showed a strong relationship between searches for ‘bitcoin’ and the volume of trading (Bitcoin to and from US dollars). Notably, the study did not report any predictive relationship with the exchange rate or number of blockchain transactions or other important statistics.
They conclude that “Google Trends is a good predictor”, at least in that the volume queries about “bitcoin” predict the volume of trading.
It’s not clear to me why we want to predict this particular statistic, unless we run a trading exchange. And in any case, we know for a fact that Google queries do not actually cause such trades, so what do we really know from this correlation?
It is pretty well understood that Google Trends reflects what the Internet is chattering about. For instance, as Marina et al note, there is a spike in capital flight from Greece, which motivated people to learn about using bitcoin to move assets, and then was followed by increased trading. We don’t really need Google trends to understand this spike in traffic, and it actually reflects the way Google is used more than the way Bitcoin works.
As Grace Caffyn notes in CoinDesk, other studies indicate a reciprocal relationship, with changes in Bitcoin activity triggering both media reports and Google searches, as well as the other way around.
Besides the basic doubts about the usefulness of these findings, there are plenty of other methodological issues. For starters, the search term “bitcoin” is pretty vague—I’m surprised they got any results at all. (Imagine using the GFT data for searches on “dollar”.) I suspect that most of the trend is coming from new users searching on the equivalent of “what is bitcoin?” or “how do I use bitcoin?”. It would be interesting to try some more refined queries like these to see if they are even better predictors.
I have to be a bit cautious about the query data itself, which is normalized. I’m not sure how large the actual volumes of queries actually are, or how much they vary.
I’m not really certain what the trading volume data means. What trading is included in this measure, and what is left out? Does this include the whole world including China? Does this include dark markets? For that matter, what is the relationship between this trading volume and the actual overall use of Bitcoin?
I’m particularly concerned because a lot of Bitcoin trading is algorithmic, so it is difficult to see how Google searches should have any connection to the behavior of the robots.
My own interpretation of these findings is that Google is one of the entry ports for people who have never used Bitcoin who want to start using it. If every new user sets up a wallet and buys a little Bitcoin (about 3 days later), then there is your correlation. Perhaps the robots are amplifying the signal, following the market trends.
In the end, I can’t get excited about this finding. A meaningless correlation between two pointless statistics.
- Martina Matta, laria Lunesu, and Michele Marchesi, The Predictor Impact of Web Search Media On Bitcoin Trading Volumes, in Information Filtering and Retrieval – DART 2015. 2015: Lisbon.