How JXEmpire's AI works
Every few minutes JXEmpire asks one question for 23 markets: will the price be higher or lower 24 hours from now? This page explains, without jargon, where the data comes from, which AI models answer that question, how the AI keeps learning, and how we check every answer.
- 23markets watched around the clock
- 3AI models give an independent forecast
- 31,079predictions checked against the real price, 48.9% correct
From raw prices to a signal, in five steps
- 1
Collect live prices
Prices, volume and the full price history of 23 markets stream in from Binance (crypto) and Yahoo Finance (stocks, indices, gold, silver, oil and forex).
- 2
Read the news
Every 10 minutes new financial headlines are scored by FinBERT, an AI that reads text and judges whether news is good or bad for a market.
- 3
Three models forecast
Three independent models each estimate the chance that the price is higher in 24 hours. Like asking three experts for their opinion.
- 4
One signal and a plan
Their answers are combined into one probability, turned into STRONG BUY to STRONG SELL, with an entry, two targets and a stop-loss.
- 5
Check after 24 hours
Exactly 24 hours later the real price is compared with the prediction. Right or wrong is stored and shown publicly as accuracy.
Step 5 feeds back into step 3: the results decide how much each model is trusted next time.
Meet the three models
No single model is right all the time, so JXEmpire uses three that look at the market in different ways and lets their track record decide who gets the biggest vote.
Technical ensemble
Rules · no trainingWhat it does. Classic trading indicators — trend (EMA 21/50), momentum, RSI, MACD, Bollinger bands and volatility — plus the news score.
How it learns. Does not learn by itself; it applies the same proven rules every time. Its say in the final answer still adapts to how often it is right.
Chronos-Bolt
Pre-trained by AmazonWhat it does. A neural network built for forecasting time series. Amazon trained it on a very large collection of real-world series, so it recognises shapes in a price chart.
How it learns. Already trained — we use it as is (“zero-shot”). It looks at the last weeks of each chart and forecasts a range for the next 24 hours.
LightGBM
Trained every night on our serverWhat it does. A fast machine-learning model made of hundreds of small decision trees. It studies thousands of past moments across all markets and what happened 24 hours later.
How it learns. Last trained 19 h ago on 15,492 examples, scoring 56.0% on data it had never seen.
Plus FinBERT, a language model pre-trained on financial text, which turns news headlines into a bullish or bearish score that the technical ensemble uses.
Is the AI actually trained? Yes — in two ways
Nightly training
Every night LightGBM is trained again from scratch on the latest price history of all markets. It sees thousands of past situations — “the chart looked like this, and 24 hours later the price was higher” — and learns which patterns matter. Part of the data is held back as an exam, so we know how it does on situations it has never seen.
Self-learning weights
Every prediction stores each model's own answer. After 24 hours we know who was right. A model that keeps beating 50% earns a bigger vote in the final signal; a model that slips gets less. The bars above show today's votes — they change by themselves as results come in.
How a probability becomes a signal
The combined model gives one number: the chance that the price is higher in 24 hours. That number decides the signal.
Example: 63% chance of UP → STRONG BUY. 50% → HOLD, because there is no clear edge.
What is in a trade plan
A zone near the current price, adjusted to recent support and resistance.
Two targets based on how far the market normally moves in a day.
The level where the idea is wrong, so a loss stays small.
How much you can win compared with what you risk.
What the AI does not do
- It does not guarantee profit. A 60% chance still means the price goes the other way 4 times out of 10. Always use the stop-loss.
- It does not use your personal data. Only public prices and news are used for training — never your account, activity or payments.
- It does not hide its mistakes. Every prediction is checked and counted, including the wrong ones.
Frequently asked questions
Is the AI really trained, or is it just rules?
Both. The technical ensemble is a set of fixed trading rules. Chronos-Bolt is a neural network that Amazon pre-trained on a very large collection of time series. LightGBM is trained from scratch on our own server every night, on the price history of all markets. On top of that, the ensemble weights adjust themselves to each model's measured accuracy.
What does the AI learn from?
Only public market data — prices, volume and volatility of each market — and public news headlines. It learns which patterns in the past were followed by a higher or lower price 24 hours later.
Is my personal data used to train the AI?
No. Your account, your activity and your payments are never used for training. The models only see public market data and news.
How accurate is it?
For a 24-hour horizon a sustained accuracy of 52–60% is realistic and valuable — markets are noisy and nobody predicts them perfectly. We publish the measured accuracy of every market and every model, so you can judge the AI on facts.
Do you also predict a week ahead?
Yes. Next to the 24-hour prediction every market gets a 7-day outlook built on daily prices and a 7-day Chronos-Bolt forecast: the chance the price is higher a week from now, a weekly signal and the likely price range. Weekly predictions are graded after 7 days (5 trading days for stocks, gold and forex) and have their own accuracy score. A week is harder to call than a day, so expect lower certainty.
Why does the signal sometimes say HOLD?
When the probability is between 46% and 54% the models do not agree strongly enough on a direction. HOLD means: no clear edge right now, wait for a better setup.
How often do predictions update?
Prices stream continuously; predictions are recalculated every few minutes and the AI market brief is refreshed regularly. Stock markets update during their trading hours.
Can I lose money following the signals?
Yes. A signal is a probability, not a certainty. Always use the stop-loss in the trade plan, risk only a small part of your capital per trade, and never trade money you cannot afford to lose.
Explore
Predictions are statistical estimates, not financial advice. Trading involves risk of loss.