• Photo: iStock

    Full Screen
  • The team's algorithm allowed for increasing profit (black) relative to the price of Bitcoin (blue).

    The team's algorithm allowed for increasing profit (black) relative to the price of Bitcoin (blue).

    Courtesy of the researchers

    Full Screen

MIT computer scientists can predict the price of Bitcoin

CSAIL/LIDS team's algorithm doubles initial investment in under two months.

Press Contact

Adam Conner-Simons
Email: aconner@csail.mit.edu
Phone: 617-324-9135
MIT Computer Science & Artificial Intelligence Lab

Scientists have crunched data to predict crime, hospital visits, and government uprisings — so why not the price of Bitcoin?

A researcher at MIT’s Computer Science and Artificial Intelligence Laboratory and the Laboratory for Information and Decision Systems recently developed a machine-learning algorithm that can predict the price of the infamously volatile cryptocurrency Bitcoin, allowing his team to nearly double its investment over a period of 50 days.

Earlier this year, principal investigator 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,” they trained an algorithm to automatically identify patterns from the data, which they used to predict prices, and trade accordingly.

Specifically, every two seconds they predicted the average price movement over the following 10 seconds. If the price movement was higher than a certain threshold, they bought a Bitcoin; if it was lower than the opposite threshold, they sold one; and if it was in-between, they did nothing.

Over 50 days, the team’s 2,872 trades gave them an 89 percent return on investment with a Sharpe ratio (measure of return relative to the amount of risk) of 4.1.

The team’s paper was published this month at the 2014 Allerton Conference on Communication, Control, and Computing.

“We developed this method of latent-source modeling, which hinges on the notion that things only happen in a few different ways,” says Shah, who previously used the approach to predict Twitter trending topics. “Instead of making subjective assumptions about the shape of patterns, we simply take the historical data and plug it into our predictive model to see what emerges.”

Shah says he was drawn to Bitcoin because of its vast swath of free data, as well as its sizable user base of high-frequency traders.

“We needed publicly available data, in large quantities and at an extremely fine scale,” says Shah, the Jamieson Career Development Associate Professor of Electrical Engineering and Computer Science. “We were also intrigued by the challenge of predicting a currency that has seen its prices see-saw regularly in the last few years.”

In the future, Shah says he is interested in expanding the scale of the data collection to further hone the effectiveness of his algorithm.

“Can we explain the price variation in terms of factors related to the human world? We have not spent a lot of time doing that,” Shah says, before adding with a laugh, “But I can show you it works. Give me your money and I’d be happy to invest it for you.”

When Shah published his Twitter study in 2012, some academics wondered whether his approach could work for stock prices. With the Bitcoin research complete, he says he now feels confident modeling virtually any quantity that varies over time — including, he says half-jokingly, the validity of astrology predictions.

If nothing else, the findings demonstrate Shah’s belief that, more often than not, what gets in the way of our predictive powers are our preconceived notions of what patterns will pop up.

“When you get down to it,” he says, “you really should be letting the data decide.”

Topics: Computer Science and Artificial Intelligence Laboratory (CSAIL), Laboratory for Information and Decision Systems (LIDS), Machine learning, Investing, Electrical Engineering & Computer Science (eecs), School of Engineering, Algorithms, Computer science and technology, Big data


How do I give you my money? And how will it get it back if it is a success?

If you want to give Bitcoin a try without spending money, have a look at http://free-bitcoin.neocities....

I'm happy to give you my money to invest haha

Github repository? Please? :)

uhh. renaissance capital

Bitcoin is unpredictable,crime however can be predicted along with all the other variables you mentioned because they have been around much longer than bit coin and we at least have a understanding of where it came from. We still don't even know who configured bit coin which is problematic because for all we know someone is in the system itself. Think about it,whoever made this technology knows the inns and outs.

Sweet. We might not be able to remove Bitcoin volatility at this point, but perhaps having it be predictable will remove the concern over volatility, or lead to more stability as more players enter the market.

what a joke.

Were exactly do you get data from "all major Bitcoin exchanges, every second for five months"?

So what now? BUY BUY BUY

Interesting !


These guys are full of it. The correlation they have found won't last; they never do when dealing with the future. Pretty soon they will be telling us that they drive to work by only looking in their rear view mirror. It will work until the big truck behind them loses its brakes. Oops, hard to predict that.

This kind of trading is good for price stability over the short term. Let's hope it becomes more widely adopted.

If you take part in an experiment then you affect the outcome of the experiment. If you observe how cattles graze and the environment that predicts their habits is one thing. But to graze with the cattles is another. By taking part in the experiment the outcome over time is not predictable. You add an element to the unpredictability.

There is indeed nothing new with this approach: I used it back in 1990 when trading on the interest rate futures for a French bank, then a Houston-based Commodity trading advisor. The underlying model was the predator/prey Lotka-Volterra model. The returns were good.

Sounds like another great innovation for bitcoin advocates.

so should I buy or sell? :p

Am I missing something? It appears these guys didn't actually trade anything. So these results mean nothing.

2x over 2 months. Not bad. My algorithm did 80x over 6 months and I'm an undergrad working alone...

sounds highly suspicious to me. is this 89% _after_ paying the spread, commission and slippage for each trade?

Bayesian regression was used 25 years ago to predict stock returns with no great success. The problem is data-mining bias, which the authors do not address. They select the best performing models without a correction for multiple comparisons. Obviously, the best model did well but many other models failed. The problem is which model to use forward. Add friction and you get a negative result.

Let the algorithm wars commence! If you want to make money, find an unsophisticated place and start a smart war. The algorithm that evolves the quickest wins.

Long-Term Capital Management rev 2.0. Taleb has written extensively on the errors of this general approach.

there needs to be interpretive algorithms that can put out false solutions to hid real ones that include algorithms created by newly creative math. this would make the block chain seem like Morse code encryption! they would need quantum encryption, but the tech. types really understand very little in the quantum realm. they crow about solutions, but what can you determine with a 140 mile limit on quantum communication with a 35 second duration or google wanting to store data on diamond based hard drives and move it around on barges. they truly do not understand the nature of the subject they are trying to understand. only a few scientist on the planet even fathom the ramification involved in this surface scratching multiverse investigate tool they are trying to tinker together!

Interesting how people are attracted to the idea of getting money without producing anything of value.

This is absolutely ridiculous, for both MIT news and for CSAIL... If you give me the ton of money that those guys got to produce such a crappy paper, I will debunk it and explain everything that is wrong with this...

Sorry, am I missing something here?

From looking at the graph on page 5 of the original paper, it seems like the researchers profit was 89% over the same time period that the bitcoin price rose over 100%. They would have done better to simply buy at the start and sell at the end.

Again, am I missing something?

Overfit, forward looking, and other leaks could explain this result

here what sort of bayesian technique is used ? naive or other ?

This is all over fitted. These sort of papers are worth less than the paper to print them. It is a shame that prestigious institutions such as MIT allow to publish such a rubbish. I am in the business of automated trading since years and papers showing such results are seldom reproducible. Authors claiming such a performance should either a) provide access to code and data to make their experiments reproducible (after all this is science), or b) stop working in academia and start their hedge fund. Those not acting according to a) or b) are charlatans, see http://www.ams.org/notices/201...

Note that someone tried to reproduce the results of the paper here: https://github.com/panditanvit...

Of course, the stellar performance could not be matched (it was even negative), and it turn out that the important clusters/features were "picked up by hand" by the authors, what a joke... The heavy math needs the manually picked-up clusters to work...

Back to the top