Welcome to PredictEvery, the home of the prediction algorithm called NonObvious.
Essentially, we have a software algorithm which takes in binary data - for instance, "1001101101101" - and outputs a binary number (0 or 1). Because the input data is always binary data, it can find patterns in that data to predict future data.
In 2007, the developer was trying to figure out how to win at roulette. In particular, even or odd (2 outcomes). Then he watched an episode of Seinfeld ("The Opposite") where George does everything the wrong way round with great success. The developer figured out that the same idea could work for roulette, because what you least expect is the next result.
Coding up the algorithm in the computer language Lisp, he discovered that it did indeed work, and not just for roulette.
After that, it was a matter of improving it in some ways.
The development has been steady over the last 2 years, with about 8 major features invented to improve things.
There must be 2 outcomes in your line of data: '1' or '0' (eg, data 011010101011111000100100101001011).
To gather the data to predict with, the far left of the line of data is the oldest result (eg, June 3rd 2012). The far right is the newest result (eg, August 4th 2014).
Here are examples with more than 2 outcomes and thus unpredictable:
Here are examples with perfectly 2 outcomes and predictable by PredictEvery:
If you can fit the thing you want to predict to 2 outcomes, then you can take full advantage of the PredictEvery software and get great predictions.
Although 16 items is the minimum, typically you're talking 100 data items. For instance, if trading gold, 100 items is 5 months (1 month is 20 weekdays). So you need about 2-5 months (between 40 and 100 results) of data for a more accurate prediction.
Typically, if the 'Overall Accuracy' is 40% or more, you can be sure that the prediction will be about 40% reliable in its prediction. And the 'How It Performed Over Time' should be roughly a diagonal line up and right.
Over 140 items of data, we can get on average 40% over that data.
If you own a predictable instrument (stocks, cryptocurrency, sport, commodities) long-term, and it doubles in value, you can't sell it and buy it back at double. But if you're day trading, you sell up every day, and so you can double the next stake to include yesterday's profits, and thus make a lot more money.
So, every 100% we make, we double the stake.
To calculate, if we make $3600 a day and the stake is 1 Bitcoin valued at $36,000, that's making the stake back in (approximately) 10 days.
If we double up every 10 days, that's a lot of money in a year: 2^30 which is off the charts and close to a billion dollar return...
As a note, because it's not necessary, if you win 20% and increase your stake by 20%, you make 1.7 times more than if you'd doubled up at 100%. So a profit of 10% becomes a profit of 17%. This is proven with a simulation we use.
Someone asked if the prediction software can cope with transaction fees.
If you play roulette, the 'fees' are 2.6% every spin.
But finance is different. Due to automation, the 'spread' is tiny.
For oil trading, the margin (or 'fees') is rarely 3% of the winnings, so a $100 win makes $3 in fees, which is tiny.
Using PredictEvery's algorithm, past performance does not guarantee future results - though it does try!