Methodology

1. The data

Every day we fetch the prices of more than 11,541 assets across 16 markets (CAC 40, SBF 120, Euronext Growth, SMI, FTSE MIB (Italy), FTSE 100 (United Kingdom), DAX (Germany), Nikkei (Japan), S&P 500, NASDAQ Growth, US large, Commodities, Commodity ETCs, Cryptocurrencies, Exchange-traded funds (ETF), Indices). To those we add the companies' fundamental data (valuation, profitability, debt, growth) and a sentiment signal drawn from the news. Prices are kept over a rolling window of about 200 days, enough to compute every indicator.

2. The technical indicators

From the prices we compute well-established indicators, grouped by family:

3. The technical score (0 to 100)

These indicators combine into a technical score. The main sources of points:

• Trend (moving averages lined up upward): up to 25 points.
• Momentum (RSI in a favourable zone, MACD positive): up to 25 points.
• Volume and accumulation (sustained volume, rising OBV): up to 20 points.
• Ichimoku (where the price sits relative to the cloud): up to 15 points.
• Market regime: a bonus or a penalty depending on trend strength and volatility.

4. The fundamental score (0 to 100)

For shares, a second score measures the company's soundness (commodities and crypto have no fundamentals, they rely on the technical score alone):

• Valuation, price against earnings (P/E): up to 30 points.
• Return on equity (ROE): up to 25 points.
• Price against book value (P/B): up to 20 points.
• Debt (borrowings against equity): up to 15 points.
• Revenue growth: up to 10 points.

5. The composite score and the signal

The score shown combines, by default, 65% technical and 35% fundamental. You can adjust that weighting (from 100% technical to 100% fundamental) to suit your style. The score reads as a plain signal:

• Buy : a favourable setup on most criteria.
• Watch : mixed signals, worth following without rushing.
• Avoid : an unfavourable setup at the time of the analysis.

Clicking the score opens the detail, indicator by indicator, with each one's contribution and historical success statistics by score band.

A stock already rated Buy or Strong Buy can gain or lose a few points depending on the tone of recent news (up to 5 points) and of social media (up to 3 points). This adjustment never applies to other stocks: it cannot turn a neutral stock into a buy.

What has changed

Label thresholds are calibrated per market: over a year of verified sessions, the average split aims at about 5% Strong Buy, 20% Buy, 35% Neutral and 40% Avoid. It is an average, not a daily quota: a falling day can show more Avoid. A calibration is only applied if, over the history, the labels stay in the right order.

Expected average split (Strong Buy / Buy / Neutral / Avoid), per market, with the calibration date:

• US (27 Sep 2026): 5% / 21% / 34% / 40%
• EUROPE (5 Oct 2026): 5% / 20% / 35% / 40%
• JAPAN (27 Sep 2026): 5% / 21% / 34% / 40%

Cryptocurrencies and commodities, too few for a calibration of their own, follow the thresholds of US equities.

A complement from a learning model can add or remove a few points to the score, market by market, after an observation period with no effect on scores. It is switched off automatically if it stops behaving as observed.

• US: under observation, no effect on the score.
• EUROPE: under observation, no effect on the score.

6. The learning model

A machine-learning model, in fact an ensemble of three algorithms (Random Forest, XGBoost and LightGBM), is retrained regularly on past data. It estimates a probability that the asset will rise in the short term, and nudges the score up or down. It expresses a probability, never a certainty, and it does not have the last word on the decision.

7. High-growth detection

A separate score spots high-growth companies that combine rising revenue, accumulation on volume and favourable momentum, without excessive debt. It only fires when every one of those criteria is met, and stays empty otherwise.

8. Sector rotation

Beyond each asset, we follow how sectors move using a relative rotation graph (RRG): which sectors lead, which are running out of steam, which way the cycle turns. That reading helps put a holding back in the context of its sector.

9. Validation by backtest

Our rules are not asserted, they are tested against past data. The backtest simulates a simple strategy to measure what it would have given: entry on the highest scores, exit at a profit target, at a time limit, or on a protective stop computed from volatility (ATR).

The results are shown as they are, including when a signal does not keep its promises. What matters is not being right every time, but that the average gain per position stays positive.

Limits and warning

Auralfa provides analysis and decision tools, not personalised investment advice. The scores, signals and backtests are provided for information. A backtest measures what would have happened, not what will happen; past performance is no guide to future performance.

The decision, and the risk, remain yours. Investing carries a risk of losing capital.