The folly of sentiment.
TLDR: Bitcoin had a wild last year and many promises for 2021. Here we analyze all the talk about bitcoin on Twitter focusing on bullish or bearish sentiment. We see that sentiment, both negative (bearish) or positive (bullish) reacts (follows) to price changes, and hint at the possibility of using sentiment as a gauge to measure the strength of a trend.
The last year was special for bitcoin, it suffered a major crash in March which it recovered quickly. The halving, as expected, happened and reduced miners reward, and last but not least bitcoin has surpassed its all-time high and continued beyond. The pandemic still hits the whole world, the global economy has seen better days but for the bitcoin hodlers and enthusiasts the first and foremost cryptocurrency is very much alive and will have a bright future in 2021.
Here at Heimdall we closely watched (and read) all the Twitter fuss during the last year. To aid our charting of the bitcoin community at Twitter, we have developed a machine learning algorithm to extract sentiment from every tweet mentioning bitcoin. In this article we intend to analyze the Twitter sentiment, both positive and negative, regarding bitcoin. We will see how the sentiment reacts to the price changes, especially in moments of sharp decline (March), or an important event (halving), and what used to be the most important price level, the all-time high of 2017.
We will make use of three metrics for our analysis:
- The number of positive tweets, from now on we will call it positive sentiment.
- The number of negative tweets, henceforth called negative sentiment.
- The price of bitcoin.
First let's take a look at the whole picture of last year:
We can see now that sentiment data does not show a trend, unlike the price. It is very hard to extract information from the above graph. However, it is more likely that Twitter sentiment is a short term phenomenon. It may vary rapidly about a level (the mean, which is fixed) but it always returns to it. It may be helpful to use the 30-day moving average of the data.
Here it is easy to see how the sentiment reacts to price fluctuation. It is evident that tops and bottoms of the positive sentiment follow those of the price. In regards to the negative sentiment, we can see it gently decreasing as the moving average of the price goes up. So we can conclude, in a preliminary manner, that the sentiment follows the changes in price. It is the expected result, since, on the contrary, it would be very surprising to find that Twitter sentiment is a significant driver of bitcoin price. So we will hold true, even if not thoroughly demonstrated, that sentiment reacts to price changes. But bear in mind that some users or some particular tweets can influence the price. Recently we had a very clear event that illustrated this point. And more
Should we abandon our analysis and just forget about the sentiment? Not so fast.
Let's analyze the first major event of 2020, the crash. Between the 11th and 12th of March, bitcoin price fell sharply in resonance with the world market. One can see in the graph below that we had a peak in number of tweets, therefore increasing the number of positive and negative tweets, which had peaks too. We must be very careful when analyzing data, and with this in mind we added the mean positive and negative sentiment (orange line). Plotting the mean, we can see that when the price was at its lowest, the positive sentiment peaked, but it was very close to the mean. The same does not apply to the negative sentiment. We can see that its true value was above the mean at almost all times at this period. When many were wary and worried about bitcoin, some were buying our intending to buy at a big discount.
Let us take a closer look. Observe that right before big drops we had peaks in negative sentiment:
In the graph above it is evident that the negative sentiment is not confined to the mean, we may even say that it is not behaving as expected. It is important to point out that bearish sentiment is considered as negative, even if the outcome of a short position might be positive. One should not interpret negative sentiment as something necessarily bad. But why this is useful? Well, a trader in possession of real-time sentiment data at a time of great volatility can use it to decide if closing out his long (taking a loss) to avoid liquidation may be the best choice, or the opening of a short position. By analyzing the behavior of the negative and positive sentiment it may be possible to be better prepared against sudden price moves and maybe infer if a drop will continue or a reversal is underway.
Let us now move to another event, the halving.
On May 11th the halving happened, that is, miners now receive less bitcoin (half) for blocks mined. This kind of event, hardcoded in bitcoin, generates diverging opinions about the price after the halving. Some argue that because bitcoin is now getting scarcer, the price will then rise but not in the short term. Again the sentiment is divisive, and like before the negative sentiment is more pronounced and is followed by a sharp decline in bitcoin price.
Much more interesting is the behavior of bitcoin in the final months of the previous year.
We can see spikes in sentiment, usually due to a spike in the number of tweets, appearing as bitcoin reaches its all-time high (19660 dollars at Bitstamp). Close to the 25th of December, the negative sentiment has a peak but prices continue to rise and are followed by a rising positive sentiment.
First let's see (below) bitcoin as it approaches the ATH (all-time high) but does not cross it. Between the 30th of November and the 2nd of January, we see bitcoin touching and crossing its historical maximum, although shyly. When it begins to rise, we see the positive sentiment increasing rapidly, way above the mean, but then the rise stalls and the optimism starts to cool down. The significant rise of negative sentiment appears later just before the price drops and levels out by noon of December 1st.
In the graph above we can see the moment which bitcoin blazes through its all-time high. Again, we can observe how the peak in positive sentiment is much more relevant, while the negative sentiment is below the mean.
We have seen how the Twitter sentiment reacts to major price changes and important price levels. It is crucial to have in mind that bitcoin is a widely popular cryptocurrency, traded in various exchanges and platforms with a considerable trading volume (lots of fake volume as well). It is a big leap to say that sentiment is a driver of price, but it is much more realistic to affirm that sentiment reacts to change in price and can in theory be used to gauge how big of a drop or a rise will be happening.
The case of altcoins can be very different, especially those with low volume and small market cap. Many of them experience major rises and falls (pumps and dumps) largely due to expectations of quick profits. Analyzing their sentiment will be the focus of a next article, where we may see if the role of sentiment and opinion is more pronounced and is a bigger influence on price.
If you want to watch the Twitter fuss on several cryptocurrencies please visit heimdall.land.
