With the cryptocurrency boom (and eventual crash) of late 2010 and early 2020, Tawheed Zaman has seen numerous cryptocurrencies rise and then disappear. As they begin, there may be some hints on social media, perhaps a glimpse of the wealth of a group of people. And then - oops, gone!
"As a scientist, I began to wonder if there was a pattern to all this noise," said Zaman, a professor of operations at Yale University. "And I was thinking that this is a financially predictable model."
The key to this currency is not long-term investment goals. You catch a wave, then you leave.
He and doctoral student Khizar Qureshi observed how people were talking about the new currency via Twitter. They believe that if you measure the conversation correctly, it will be possible to identify the coins with the best odds in the next month.
"The point of this currency is not long-term investment goals," Zaman said. "You ride a wave. That's a lesson in cryptocurrency trading in general."
Zaman and Qureshi achieved their results by inventing a new ad dropper method. In the past, people have tried to use the raw volume of tweets on a particular topic to predict outcomes, assuming that many tweets show high performance in the future. But Twitter limits the amount of data that can be removed from its site, meaning the amount of raw data is sometimes too much for anyone to get a meaningful sample. It's impossible to track millions of tweets every month.
Instead, researchers examined the predictive power of sentiment analysis; Useful to discuss a specific topic? But internal shorthands like #buythedip or #hodl, both of which convey positive sentiment in the crypto world, tend to escape machine learning analysis, as do memes that express sentiment one way or the other.
Instead, Zaman creates an "engagement ratio" based on the number of followers of accounts that post tweets mentioning the cryptocurrency, as well as the number of likes and retweets per tweet. These two metrics combine to create a single number between zero and one that represents how many people are talking and hearing about that cryptocurrency in a given month. Zaman and Qureshi used this index to benchmark a sample of 48 cryptocurrencies entering the market between 2019 and 2021 and speculative investments within one month. This investment generated an (estimated) return of around 200%.
This allows the overall temperature of an object to be measured by sampling a relatively small amount of data. With just a few thousand tweets, you can test out a new cryptocurrency, movie, brand, product or politician.
"One of the really interesting things is that this signal was not monotonic," Zaman said. Researchers surprisingly found that if the participation rate of a particular coin is below a certain threshold, then that coin should not be bought. Too much noise is also a bad sign, he adds. "If the ratio is really high, you also want to avoid buying coins." The very high probability seems to be a large number of bots manipulating the currency and possible fraud where people have artificially inflated interest among buyers before the currency crashes. He said it was a Goldilocks place where the investment was worth it.
Zaman notes that this insight could be useful for regulators trying to stop fraud. When a coin is listed on a cryptocurrency exchange shortly after it is listed, it starts making an unusually high amount of noise and this can be a sign that the coin is being used.
Zaman says that the equity ratio doesn't just apply to the world of cryptocurrency; In fact, he recently tested it in another area that is very difficult to predict. During a social media class, he asked students to test whether a new method could predict a movie's performance. They collected historical promos for several films and then tried to decide which ones would work. Super Mario Bros cartoon. It was definitely the most popular movie and in fact it is now the highest grossing movie of 2023.
"It allows you to measure the overall temperature of a subject with a relatively small amount of data sampling, and it seems to be a very good predictor of success," he says. "With just a few thousand tweets, you can see a cryptocurrency or a movie, or a new brand, product or policy."