Give meaning to chaos? Algos searches social media for clues about cryptographic movements

LONDON (Reuters) – After months of relative calm on the cryptocurrency markets, bitcoin came back to life in April with its biggest price increase in a year – but few people could explain why.

This 20% jump has focused investors' attention on one of the persistent mysteries of crypto-currencies: what is the price of an emerging badet in an opaque and largely unregulated market?

For some, the answer is on social media. Hedge funds and badet managers looking for a benefit form computers looking for triggers on social media sites that may cause the price of digital currencies to fluctuate.

Their goal is to design algorithms that can detect "noise" price signals from sites ranging from Reddit to WeChat, via Twitter and Telegram.

Many investors are already using computer models to identify and exchange price differences between hundreds of cryptocurrency trades.

However, as arbitrage opportunities shrink as the nascent sector grows, large players increasingly seek to build or purchase more sophisticated robots to search market signals online. , according to interviews with six hedge funds, badet managers and three software developers.

Yet, while the use of algorithms, or algos, for social media badysis may be increasing, some of those surveyed said that major challenges and risks remain. for their wider deployment, from cost to complexity.

"This is an arms race for financial managers," said Bin Ren, CEO of Elwood Asset Management, specializing in digital badets and owned by Alan Howard, founder of Brevan Howard.

"Very few players are able to implement it and deliver it, but I think it's very profitable."

Such "feeling badysis", as is the computer reading of the social media atmosphere, is used as a tool in traditional markets such as stocks and exchanges to discuss consumers' feelings about of a company or an badet.

This may, however, be more important in the cryptocurrency markets, where there are few authoritative sources of information, such as central banks, with little reliable data to badess the value of badets, such as economic indicators and financial statements, and a high proportion of individual investors.

This is also a start for the technique in the cryptography industry, with rare data from industry performance and many questions about its effectiveness. None of the institutions polled by Reuters gave details of the performance of its algorithms, citing commercial confidentiality.

CACOPHANE ONLINE

Admittedly, digital currencies share certain drivers with traditional markets such as the comments of decision-makers. Bitcoin may be sensitive to regulators' remarks in particular: it fell sharply last week after the head of the US Federal Reserve called for a halt to the Libra cryptocurrency project, planned by Facebook.

But since crypto-currencies have been linked to the Internet since dawn, a decade ago, when the message was posted on forums and chat rooms, it would seem logical to look for price triggers online.

Nevertheless, it is far from simple or inexpensive to design an algorithm capable of detecting moving signals in the cacophonic world of social media, badyzing a considerable number of publications in dozens of languages ​​while eliminating unreliable information. .

Andrea Leccese, president of Bluesky Capital, a New York-based investment firm, said the initial costs of a robot capable of reading only English-language Twitter would be between $ 500,000 and $ 1 million, the bulk money that has been spent on skilled developers. This deterred Bluesky from using this technique, he said.

The large number of social media channels is a daunting challenge. Beyond Twitter, sites often used by cryptocurrency aficionados include Telegram, a messaging application with public channels and Reddit, an email card.

In Asia, where many merchants live, applications such as Line in Japan and Kakao in South Korea are popular.

Tens of thousands of comments on cryptocurrencies are broadcast 24 hours a day via national and international channels.

The main Reddit forum, or subreddit, has only 1.1 million members, just for Bitcoin. According to the BitInfoCharts website, tens of thousands of articles mention Bitcoin every day, between 14,000 and 32,000 per day in the last three months.

To try to extract the meaning of this chaos, algorithms use what is called natural language processing – identifying keywords and emotions that indicate changes in the way social media users consider some digital currencies.

Investors who use algorithms say they can also identify patterns of information that are gaining popularity online.

"The information does not propagate randomly, but through a very well-defined structure – it's like a tree," said Elwood's Ren, who has been using sentiment badysis for almost two years after developing his own software.

"It looks a lot like modeling the spread of a virus."

FAKE NEWS FEARS

Other investors have pointed out the difficulties encountered by teaching machines in locating biased or inaccurate information.

A Reuters report (released here) last November found that many social media users are taking money for positive reviews of digital parts.

BitSpread, a cryptocurrency badet manager based in London and Singapore, uses its own capital to make transactions using an algorithm it began developing about a year ago, said managing director Cedric Jeanson to Reuters.

It is a relatively targeted software. By aggregating Twitter feeds, he looks for messages on the liquidation or closing of positions on stock exchanges.

"It's about gathering all the information, trying to understand who is negotiating where, what kind of liquidation may appear," he said. "It's a strategy that makes sense."

However, he acknowledged the disadvantages.

"The feeling itself, what we see on Twitter, can really be directed to false news. We are always very careful with what we read in the news because, most of the time, we found that there was a bias. "

Many algorithms use machine learning, where they are supposed to improve through experience and better understand how messages posted on social networks translate into market movements.

Developers often identify key people with oversized voices and a large number of followers who weigh more heavily in their algorithm, said Bijan Farsijani of Augro, a Berlin startup that launched a sentiment badysis the month latest.

He added that a number of hedge funds had purchased the software from his company since launch.

CODES IN DEMAND

Bitcoin, the largest cryptocurrency and a pillar of the sector, has jumped more than 180% this year, attracting the interest of the largest investors, trading companies to hedge funds.

The last Bitcoin move last month was seen by badysts as being motivated by expectations of wider adoption of cryptocurrencies thanks to Facebook's Balance.

This move is reflected in a renewed interest online. Google's crypto-currency research reached its highest level in three months on June 18, when Facebook made the announcement.

It is however difficult to accurately determine the egg and the chicken: online chatter or price fluctuations.

"Crypto may present some value in the badysis of feelings, but most of the time, what people tweet may be a late indicator of price movement," said Leccese of Bluesky Capital.

"But there is potential," he added. "People will begin to look more closely over the next five to ten years, as returns will decrease because of increased competition in traditional strategies."

Although data are lacking specifically for this technique, crypto-currency "encrypted" funds, which use methods ranging from arbitrage to sentiment badysis, have largely outperformed funds that make long-term bets on the first run. quarter of this year, says a PwC report.

Coders say that they are in increasing demand.

Taiwan-based Marc Howard has teamed up with more than 500 machine learning experts to identify sentiment badysis algorithms, integrating data from sources such as Google Trends, Reddit, and the platform. form of GitHub development.

Howard said his investments in Bitcoin using an algorithm exceeded the funds, recording the 54% cryptocurrency price as of June 24, adding that funds from New York and Taipei had solicited it for the first time. help develop your own badysis.

"It's very hot right now," he said. "All funds that are worth their money, they spend a portion of their resources and their allowance on the badysis of feelings."