Using Machine Learning to Predict Virtual Currency Exchange Rate

By December 04, 2014 1 comment
Machine learning

Researchers at MIT have recently developed a machine learning algorithm that can potentially be used to predict the going rate for Bitcoin. This could be handy for those who use the crypto-currency or wish to invest in it, but the same approach could well be used to predict other kinds of trends based on historic data analysis.

Machine learning is a scientific discipline that deals with the construction and study of algorithms that can learn from data. Unlike traditional computer processing, which simply follows explicitly programmed instructions, such algorithms operate by building a model based on data inputs and using that to make predictions or decisions. The process of teaching machines to develop learning processes so that they become ‘intelligent’ comes under the overall umbrella of what is known as ‘artificial intelligence’ (AI).

The AI field is seeing a great deal of investment by the Web giants, with Google out in front. In January the Mountain View data specialist paid $400 million to acquire DeepMind Technologies, a twelve-person start-up specialising in ‘Deep Learning’. Deep Learning is an approach to artificial intelligence which seeks to mimic the neuron connections of the human brain, specifically the cerebral neocortex. Facebook has been showing interest in this technology. Facebook AI Research is looking to make major progress in machine learning and last year Mark Zuckerberg’s firm recruited New York University professor Yann LeCun, who is a specialist in the field.

Meanwhile researchers at MIT’s Computer Science and Artificial Intelligence Laboratory recently developed a machine-learning algorithm designed to predict the exchange rate of e-currency Bitcoin. When they tested it out in practice the team managed to nearly double its initial investment over a period of 50 days.

The Bitcoin system is technically a peer-to-peer payment network that operates on a cryptographic protocol – a so-called ‘public’ key which encrypts data so as to protect it from online piracy, and a ‘private’ key which authenticates the transaction. People who have signed up to Bitcoin can exchange the network’s crypto-currency. Transactions are recorded in a distributed public database based on file-sharing, known as the ‘block chain’, which guarantees consensus on the deals executed by means of a proof-of-work system called ‘mining’.

Earlier this year, MIT Principal Investigator (senior researcher) Devavrat Shah and recent graduate Kang Zhang collected price data from all major Bitcoin exchanges every second for five months, accumulating more than 200 million data points. Using a technique called ‘Bayesian regression’, which enables probabilities to be drawn from random events, they then trained an algorithm to automatically identify patterns from the data, which they used to predict prices and trade accordingly. Every two seconds the MIT Computer Science and AI specialists predicted the average price movement over the following ten seconds. If the expected price movement was higher than a certain threshold they bought a Bitcoin. If it was lower than the downside threshold they sold one; and if it worked out somewhere in between they did not trade.

As Bitcoin is still in the process of gaining general acceptance as a financial asset, and people are still trying to figure out whether it can really be called a currency or is simply a piece of technology or a basic raw material, its ‘exchange rate’, or price, is highly volatile. The Shah-Zhang algorithm demonstrates however that it is perfectly possible to speculate with the crypto-currency, which highlights that is potentially vulnerable to attacks by determined speculators. In fact Devavrat Shah had previously used the algorithm-plus-Bayesian regression approach to predict Twitter trending. He also believes it is feasible, with enough data points to draw on, to apply the technique to explain Bitcoin (or other) price variations in terms of specific factors relating to the human world.



Page top

1 Comment


Submitted by cosmo miller (not verified) - on December 04, 2014 at 11:10 pm

Legal mentions © L’Atelier BNP Paribas