The Corona Crisis and Bitcoin Attention
The corona crisis has dramatically changed the world we live in. It has led to drastic changes in our personal lives and the economy. Bitcoin should be designed to thrive in these uncertain times and one would expect interest in bitcoin to rise. In this article, I am looking at how the corona crisis has influenced bitcoin attention using GoogleTrends data. The focus is on Google search data for the US using daily data from 01/01/2020 to 31/08/2020 (for a detailed data description see Appendix).
Before getting into the more detailed analysis let us first have a look at how bitcoin search intensity and corona search intensity have developed over time. In Graph 1 we see how bitcoin search intensity has changed over this year. Over the course of the year, relatively high volatility of bitcoin searches can be observed with two large spikes in the first half of the year. The first spike in March seems to be associated with an increase in corona search interest and both seem to move in tandem above a longer time period. The second spike in bitcoin interest coincides with the block reward halving of bitcoin where the block rewards for miners did halve from 12.5 BTC to 6.25 BTC on May 11, 2020. As can be seen, bitcoin searches and halving searches went up sharply in the weeks before the halving event. At the first glance, both search terms, corona and halving, seem to be correlated with bitcoin searches. However, a lot of things have happened during the course of this year and other factors are potentially at play as well. One other important factor seems to be the unprecedented monetary intervention of central banks in the economy and financial markets.
European Central Bank (ECB) and Federal Reserve Bank (FED) central bank search intensity show a relation to bitcoin search intensity (see Graph 2). The relationship is most visible in some of the spikes. The searches for ECB and the FED are highly volatile though, which makes it hard to clearly see that relationship by purely looking at the graph. The halving event, the corona crisis and monetary policy are only three major factors that might have influenced bitcoin searches. Additionally, there are other search terms and factors that potentially have played an important role in driving bitcoin interest during the course of the corona crisis (like e.g. gold searches or searches for the USD).
With all these factors at play and some of these happening at the same time it is hard to get an understanding of what is actually driving bitcoin searches by just looking at the raw data. To get a better understanding of the correlation of bitcoin searches and other searches in the US linear regression models are estimated to control for a set of Google search terms. The results of the estimations are reported in Table 1 below (to make the interpretation easier the significant results have been marked in green).
The first column shows the estimation where only the relationship between bitcoin searches and corona searches is estimated. The relationship between both is positive and significant. However, as mentioned earlier a lot of things have happened this year so we don’t know if potentially another factor is causing this rise in bitcoin searches. To see how the relationship between the two behaves and whether it is robust to changes in the estimation I am adding more search terms with each regression. Irrespective of which control variables are added to the estimation corona searches are positively associated with bitcoin searches. So the corona crisis is indeed associated with higher bitcoin searches.
Estimation 6, 7 and 8 are the most relevant ones because here all the significant variables are included. According to estimation 6 a change in corona searches by one point is associated with a change of bitcoin searches of approximately 0.21 points (in the following for the effect sizes I will refer to the results of estimation 6).
For bitcoin-related search terms I find a positive relationship for two of the included variables. First, for halving searches which are in comparison to the other coefficients strongly related to bitcoin searches (an increase in halving searches is correlated with an increase in bitcoin searches of roughly 0.69). Second, searches related to the stock to flow model made prominent by PlanB are positively associated with bitcoin searches. The correlation between stock to flow and bitcoin searches however is relatively low (an increase in stock to flow searches by one point is correlated with an increase in bitcoin searches of about 0.04 points). Remarkable is that gold searches are robustly related to bitcoin searches throughout the estimations. Here the similarities of bitcoin to gold seem to be at play. The relationship is also relatively strong with 0.66 points. This is a strong indication, that bitcoin is more and more perceived as digital gold in the US. Two other indicators indicating that bitcoin is perceived as digital gold and might be considered as a safe haven are the two central bank searches related to the ECB and the FED. Both are positively associated with bitcoin searches. Last but not least searches for the USD are significantly and strongly related to bitcoin searches. In fact, USD searches show the strongest relationship of all of the search terms. For instance, if we take the coefficient of estimation number 6 we can see that an increase in the search index for USD by one point is correlated with an increase in bitcoin searches of roughly 6.80 points. This is roughly ten times higher than the correlation of the halving search term with bitcoin searches. That the correlation is so much higher for USD searches as opposed to the other search terms might be for two reasons. First, the other search terms are rather specific and thus might not be searched as much by the broad population. These are the bitcoin specific terms (halving, digital gold, and stock to flow) which are likely mostly searched once people are already into bitcoin already and then the financial terms (gold, ECB, Fed). As opposed to these search terms changes in the value of the USD directly impact the lives of people and the effects of changes in its value are potentially more easily understood than e.g. an increase in quantitative easing by the central bank. This leads us to the second reason which is the value of the USD. In Graph 3 the exchange rate of the USD against the Euro over time is depicted. At the beginning of the year the USD has appreciated against the Euro, with very volatile periods. But from May onwards the value of the USD against the Euro started to drop strongly. Over the whole study period the USD/EUR exchange rate has declined from 0.8919 to 0.8377. That is a decline of about 6%. What is possibly happening here is, that US citizens start considering bitcoin as a way to protect their savings against the devaluation of the USD. If we take that as given this means that a weak/depreciating dollar is positively linked to bitcoin search interest.
In summary the coronavirus and searches related to the coronavirus did indeed play a role in bitcoin search interests. Overall the results are supportive of the current sentiment that bitcoin is thriving in uncertain times and is more and more manifesting its place as digital gold and a safe haven and is used to store value in reaction to increasing monetary policy intervention by central banks and changes in the USD value.
The data used in this analysis is retrieved from GoogleTrends using the gtrendsR package from R. GoogleTrends only allows you to retrieve daily search data for up to 9 months. That is why I have chosen the period from 01/01/2020 to 31/08/2020 to have daily search data. GoogleTrends reports the search intensity as an Index from 0 to 100. For weekly data the index reaches the value 100 on the day the highest search intensity occurred during the reported period. GoogleTrends allows us to retrieve search data for up to five search terms at once. When simultaneously retrieving five search terms these search terms are weighted against each other. This allows us to see which search terms have been more important during the study period. It also allows me to create an index of these terms when needed. For some of the variables used in the analysis an index of multiple variables has been created (with a maximum of five variables included). The variables and the composition of the variables are shown in Table 2 below.
The index is calculated as follows:
First, the mean for all the search terms that are included in the index across the observation period is calculated. Afterwards the weight for a search term is calculated as follows (exemplary for search term 1 of the index):
The final index is calculated as shown below:
t=time; n=number of search terms included in the index