MIT researchers have managed to create an algorithm that predicts the price of Bitcoin, offering a reliability of 4 to 1 in the investment.
The oscillations that the price of Bitcoin has experienced in recent months and practically since its creation as a virtual currency have been at the same time a fertilizer for speculation and a source of controversy. Of the differences in value over time are good proof cases such as that of Kristoffer Koch, a university student who bought 27 dollars in Bitcoin to investigate the subject of his thesis, data encryption. This was in 2009. Four years later, forgotten that Bitcoin purchase, he looked back at his account and he found that those $ 27 were now worth $ 886,000. This story certainly encouraged some to invest in virtual currency, but at that point it was no longer the same.
For those who have the crazy hope of having an experience like Koch’s, a team of scientists at MIT has created an algorithm capable of predicting the price of Bitcoin. For better or for worse oscillations could be foreseen thanks to an artificial intelligence system, where the ‘machine learning ‘ factor is key.
Researcher Devavrat Shah, a specialist in electrical and computer engineering, has developed with Kang Zhang, a recent graduate student, a method that applied to the world of Bitcoin has allowed him to practically double the amount invested in virtual currency. In the experiment that both carried out the results have been monitored for 50 days, a period in which earnings have increased by 89%.
The algorithm is based on a Bayesian regression model and is able to identify patterns in all the information it processes. This analysis is then used to detect price trends and predict whether the price will rise or fall. Every two seconds the algorithm launches a new prediction for the next 10 seconds, which means that it is constantly updated. The researchers set a threshold to invest, if the predictions exceeded it they reversed.
In this way they got a security of 4 to 1 in their investments, where 4 is the investment returned for every 1 risk opportunity. Needless to say, the algorithm designed by the scientists at MIT has a specific job, apart from the machine learning it has carried out while it was in operation. For five months, prior to the 50-day trial, the system has been collecting data every second of Bitcoin price and transactions that took place in this currency.
In total, the algorithm had more than 200 million information points. With regard to the 50 days of operation, to achieve this 89% of profits have had to be made 2.872 transactionss. So if someone was thinking of using this system to buy Bitcoin manually can go forgetting. Instead, this kind of experiments push a little more the limits of artificial intelligence and especially the field of ‘machine learning‘, one of those that has the greatest projection to solve problems in the future.
Images: zcopley and Whitez
