While many have tried to formulaically or algorithmically derive ways to predict daily prices of top cryptocurrencies, for example the price of bitcoin in 2019, there haven’t been any tools for intra-day projections. Further, many if not most of these daily predictions have been steeped more in belief than in fact.
Over the last year and a half The TIE has researched intra-day cryptocurrency price movements and sought to identify tools which offer evidence-based explanations for those changes.
What our research has shown is that sentiment analysis is a powerful tool for explaining and projecting future asset prices, and in particular cryptocurrencies. Partnering with Social Market Analytics (SMA), a market leader in the provision of proprietary predictive sentiment analytics to quantitative hedge funds and large financial firms including Fidelity and Cboe, The TIE has researched the correlations between changes to sentiment of over 250 of the leading cryptocurrencies and respective future price movements. Leveraging access to the entire live Twitter feed, the Twitter Firehose; sentiment is calculated in a patented three step process: extraction, evaluation and calculation. The Twitter firehose is scanned to extract all tweets relevant to an entity. The tweets are tagged with a sentiment from a natural language processor that uses machine learning techniques to identify sentiment as it is related to finance. The source of tweets is evaluated and all duplicate and spam content is removed. The filtered stream is aggregated to calculate a quantifiable sentiment. All of this is done in less than 1/3rd of a second.
The TIE quantifies this sentiment in the form of a hourly sentiment score ranging between 0 (Very Low) and 100 (Very High). This score is based off of Twitter conversation on a particular cryptocurrency over the last twenty-four hours, heavily weighted on the last hour, and is calculated in real-time and updated on The TIE every minute. For more information on how sentiment is calculated, why we use twitter, how we deal with bots, how we score individual accounts, and other frequently asked questions click here.
The correlation between a cryptocurrency’s sentiment and price is then calculated. This score ranges from -1 (perfect negative relationship) to 1 (perfect positive relationship). While in most cases there is a positive correlation between a cryptocurrency’s sentiment and its price, we have identified many instances in which cryptocurrency sentiment is inversely related to price movement. Three examples include Metal, Ambrosus, and Genesis Vision which have historically shared this negative relationship.
Twitter is a global platform with over 850 million tweets sent daily. The conciseness of the micro blog ensures that correct information is conveyed most efficiently. Our research has shown that when using our patented algorithms to remove noise, sentiment has predictive power.
Leveraging our sentiment engine, The TIE is able to pick up those instances in which extremely positive or negative conversations are occurring around a particular cryptocurrency. In many instances, that negative or positive conversation is driven by news, which breaks on Twitter first. An example of this was the news of the Bitcoin ETF being rejected on July 26th. As soon as the ETF was rejected and the news hit twitter, The TIE immediately picked up negative sentiment on Bitcoin. That sentiment quickly dropped extremely negative and downward price movement trailed for 15–30 minutes following the negative sentiment spike. As sentiment began to neutralize and turn positive, upwards price movement followed.
Our research has shown that sentiment consistently helps traders generate alpha, both on the long and short sides of the market. To demonstrate this we created two indices. A baseline index and a Sentiment RSI Long/Short index. In both instances the top 20 cryptocurrencies by market cap were selected. In the baseline strategy we equally weighted each cryptocurrency 5% and rebalanced them daily. In the long-short strategy, the same cryptocurrencies were balanced daily based off of sentiment momentum. Coins with positive sentiment momentum were purchased and those with negative sentiment momentum were shorted. The TIE’s long-short strategy achieved a cumulative return of 29.31% between January 1st and June 25th 2018, while the baseline composed of the same coins achieved negative returns of -67.91%.
We then took the same basket of twenty coins and looked at their performance in three strategies. In the base portfolio we equally weighted each cryptocurrency 5% and rebalanced them daily. In the tilt portfolio, the same cryptocurrencies were balanced daily based off of positive sentiment and price momentum. In the tilt underweight strategy, the same coins were selected and balanced daily based off which coins had the worst sentiment and price momentum. As expected, coins with positive sentiment and price momentum substantially outperformed the market, while those with negative sentiment and price momentum underperformed.
We then wanted to test the effectiveness of trading off of just sentiment and how a purely sentiment driven strategy would have performed during the bull run of late 2017 and early 2018. We grouped a universe of the top 147 coins into quintiles based off of their sentiment.
Our research found that purchasing coins in the top quintile for sentiment would have produced a return of 904% between July 2017 and February 2018 compared to the broader market return of 720%. Similarly coins that ranked in the top 60–80% for sentiment achieved a return of 815% over that same time span, outperforming the universe of tokens. On the opposite end of the spectrum tokens in the bottom quintile achieved a return of 640% and those in the bottom 20–40% returned 605%, below the return of the overall crypto universe.
To convert raw sentiment scores to an actionable multi-factor signal The TIE has developed 1Hr Price Projection Ranges. These ranges, calculated using 90% confidence intervals, consider the current hourly sentiment of a cryptocurrency, the volume of conversation on Twitter surrounding a particular coin, and the weekly volatility on that coin’s price.
For Bitcoin the price projection range is typically around .75%. Historically, The TIE has accurately predicted 92.76% of the time the price of Bitcoin an hour before within a small .75% range. For example if The TIE provided a 1Hr Price Projection range for Bitcoin between +2% to +2.75%, 93% of the time the price would have stayed within that range.
In the case of Civic, shown above, at 4:10PM UTC on October 2nd the actual price of CVC was $.12359. At that time The TIE sentiment engine picked up a large influx of positive tweets and provided a 1hr price projection for 5:10PM of $.13015 (+5.3%), with a range of $.12889 (+4.3%) to $.1314 (+6.3%). The actual price of Civic at 5:10PM was $.13029 (+5.4%), or just .1% off from the estimate. Beta testers who had set an alert on The TIE to receive an email or text message for each instance in which Civic had a 1hr price projection greater than 5%, were able to purchase and sell that coin within that hour and capture a 5.4% profit.
As research by The TIE has shown, sentiment is a powerful tool for analyzing the cryptocurrency market and for predicting future price movements. In an asset class void of traditional fundamental data, sentiment is the most significant driver of price movement. Cryptocurrency valuation is driven by the wisdom of the crowd, and The TIE is at the forefront of analyzing investor sentiment and building predictive models to forecast that movement.