Using Google Trends to spot market cycle tops
Looking at Google Trends charts can tell us a lot about what people are searching / thinking / doing or about to do. One of the use case I’ve been trying to develop is trying to spot market cycle tops for crypto currencies.
As one can easily see from the chart below, raw data from Google Trends is not enough to extrapolate any useful information, as it is impossible to differentiate a mid-cycle top from an end-cycle one. Here, someone who would have looked at Google Trends’s results for the word ‘Bitcoin’ back in May 2017 would have probably thought the cycle top was in. It wasn’t.
One of the main problem is that, as bitcoin adoption grows, so does price and search engine usage data (which is auto calibrated by Google to maintain the ‘min =0 ; max = 100’ rule), meaning that past data will tend to converge towards 0 as time goes by, making it impossible to compare two cycle tops).
A first way to improve the data set would be to divide Google Trends data by the BTC adoption rate, or price as we lack the former. Doing this yields the following chart:
We now have a slightly more useful chart, though not perfect yet. We could start to extract some data based on a comparison between each data point and a specific level (arbitrary level, moving average, you name it). In the following chart (log scale), I selected 75 as an arbitrary reference level to spot market exuberance while killing useless ‘data noise’. Each bar points to a date when [Google Trends data] / [BTC price] was above 75.
Interestingly, this chart only yields two data points which happen to coincide with two tops, 2 weeks in advance, including the final Bitcoin top, in mid-December 2017.
Still, two major flows remain: First, looking at ‘Bitcoin’ Google Trends data alone is not enough (lacks diversification). Second, setting the 75 level would have been more difficult back in 2017, when 2019/20 data wasn’t available.
To deal with the first point, we can easily look for other terms in Google trends such as ‘crypto’, ‘crypto wallet’, ‘HODL’ etc and combine the results to improve the overall accuracy. I like ‘HODL’ the most since people searching this in Google are likely to be inexperienced crypto users trying to understand why everyone is writing ‘hodl’ instead of ‘hold’ on Twitter. The mass arrival of inexperienced traders often matches cycle tops.
The below chart corresponds to the sum of the adjusted data sets for the words ‘Bitcoin’, ‘Crypto Wallet’, “HODL” and ‘Crypto’ while the orange line is simply the 7-week moving average.
We can now look at the spread between the sum and its moving average to spot periods of time when people rush to their search engine to find information on bitcoin or other crypto currencies, which usually tends to indicate cycle tops.
To get a more responsive data set, one can use the product of all the different data sets rather than the sum.
What I particularly like is multiplying the product by the spread (spread between the sum and its 7w moving average), while killing any noise using a Y/N filter yielding 1 or 0 depending of the spread itself (e.g. multiply the data by 0 when the spread is below a certain level and by 1 otherwise). See below.
Here, the December 2017 top would have been clearly visible as the indicator would have been flashing red during 2/3 weeks before the market top.
Though not perfect, using Google Trends data can tell us a lot regarding social trends, speculation and more importantly, about the average profile of active users at any given time (more professional traders at cycle lows ; more inexperienced traders at cycle tops). Monitoring this information closely could help you find the next bitcoin top, which I expect to be around mid-August 2022.
Please do not hesitate to comment and tell me if you have other ideas/better ways of using/adjusting Google Trends’ data. One thing I discovered building these charts is that we definitely are miles away from reaching the next cycle top. From an adoption perspective, Bitcoin’s future is bright. Do not miss out.
Follow me on Twitter: (De)CryptEd