AI powered financial assistant
Connecting finance and technology. Machine learning and AI in making sense of different data. Valuable decision making is taking over the news and events' speed.
Profile
We work on AI models processing financial, economical, news and events data with fast intellectual machine models. Economical behaviour can't always be explained directly. Our assistants work with explicit, hidden or implicit relations and can forecast.
What is risk?
An academical definition : a deviation from the expected pattern. But computer needs an algorithm. Practical approach - combination of 2 solutions :
1. Positive expectation of profits.
2. Growing capital curve smoothing.
Finally we get acceptable probabilities and yield.
Trader = Gamer?
Trader = Gamer => Decision Maker. Every decision has price and value. Btw. With right financial cost function single pc replaced IBM cluster in chess matches between Gary Kasparov and Deep Blue
Modern financier formula of knockout
1. Sufficient odds.
2. Acceptable probability.
3. Good explaining factors selection.
Partial solution tells that only probabilities of winning trades > 70 % and odds > 3 / 1 can be more or less comfortable for decision making.
Winning speculator gene
Input : Any data.
Machine : Multiple dimensions, online learning.
Output : Price move, time lag, volume.
"Buy The Rumor Sell The Fact" Modern approach
News vector representation as input and price reaction as response. Convenient odds (good move), acceptable accuracy of model and that's all.
How do not loose yourself
1. Select a trading suggestions, for example, with accuracy > 0.85%, odds > 4/1.
2. Rank it by an expectation of profits.
3. Choose one of the suggestions or use Optf criteria for a portfolio.
These principles are well known but expensive in computing, especially online semantic processing and machine learning. A digital assistant should only do all automatization.
Modern news trading approach
Modern news trading assistant should be online machine capable to rank many news/price pairs and suggest best choices. Selection can be done automatically too. But inference is always very interesting.
There are 2 silver bullets that can stop TA Witnesses
1. Two known open researches : PennLehman competition where pro trader automation vs TA has won; Machine learning by adaptive boosting author. TA is just chart painting.
2. Our favorite : show odds & probabilities.
Current projects state
We move financial AI assistants prototypes to working services. Semantic models are trained on 2 years news and events data As usual many tests before ignition.
