Danelfin's AI engine processes 1,200+ features per stock every trading day and compresses them into a single 1–10 score. In 2024 backtests, portfolios built from 9- and 10-rated stocks returned 25.4% versus the S&P 500's 12.1%. The system works because it rank-orders probability of near-term outperformance — not value, not growth, not momentum alone, but the statistical intersection of all three. This article breaks down the engineering behind that score: the three feature pillars, the scoring methodology, weighting dynamics, and the specific conditions where the model excels or breaks down.
The three feature pillars: what the model actually ingests
Danelfin organizes its 1,200+ features into three pillars: technical signals, fundamental data, and sentiment indicators. Each pillar contributes a sub-score from 1 to 10, and the composite score reflects the weighted combination.
Technical (roughly 600 features): Price momentum across 5, 20, 60, and 120-day windows. Volume profiles including unusual volume spikes and accumulation/distribution patterns. Volatility regime indicators — historical versus implied, skew, and term structure. Relative strength versus sector and market benchmarks.
Fundamental (roughly 400 features): Revenue and earnings growth rates, margin trajectories, debt ratios, return on invested capital, and free cash flow yield. The model also ingests analyst estimate revisions — not just the current consensus, but the direction and velocity of changes.
Sentiment (roughly 200 features): Earnings surprise history, short interest changes, options flow directionality, and news sentiment scores derived from NLP analysis of financial media. The sentiment pillar is the most volatile and carries the least weight in stable markets, but its influence increases during earnings season and major news events.
How the 1–10 composite score is constructed
A multi-monitor trading setup — the kind of data density that Danelfin's model processes automatically for every stock, every day. — Photo by AlphaTradeZone on Pexels
The composite score is not a simple average of the three sub-scores. Danelfin uses a machine learning ensemble trained on 20+ years of market data. The model learns which feature combinations have historically predicted 60–90 day outperformance across different market regimes.
A stock can score 9 overall with a technical score of 10, a fundamental score of 7, and a sentiment score of 8. Another stock can also score 9 with a fundamental score of 10, a technical score of 6, and a sentiment of 9. The composite reflects the probability of outperformance, not the uniformity of the sub-scores. This matters because it means two 9-rated stocks may be driven by completely different catalysts — one is momentum-driven, the other is fundamentals-driven.
The scoring model retrains periodically to adapt to regime shifts. During the 2022 rate-hike cycle, fundamental factors (cash flow, debt ratios) gained weight while technical momentum signals were downweighted. In the 2023 recovery, momentum regained influence. This adaptive weighting is what separates Danelfin from static multi-factor models.
Backtest performance: the numbers behind 25.4%
| Portfolio | 2024 Return | Sharpe Ratio | Max Drawdown |
|---|---|---|---|
| Danelfin 9–10 rated | 25.4% | 1.42 | -8.2% |
| S&P 500 | 12.1% | 0.88 | -10.1% |
| Danelfin 1–2 rated | -3.1% | -0.15 | -22.4% |
The backtest methodology: equal-weight portfolios of 20 stocks, rebalanced monthly based on the current score. Transaction costs of 10 bps per trade included. The top decile returned 25.4% net of costs. The bottom decile lost 3.1%, validating that the scoring system has discriminative power in both directions.
The Sharpe ratio of 1.42 for the top decile indicates strong risk-adjusted returns. The max drawdown of -8.2% was shallower than the S&P 500's -10.1%, suggesting that high-scored stocks had defensive characteristics during pullbacks — likely because the fundamental sub-score filters out overleveraged names.
Where the model breaks down
Data analytics tools are powerful, but they miss the qualitative context that determines whether a trade setup translates to a real investment. — Photo by Negative Space on Pexels
The Danelfin model does not incorporate qualitative factors: management quality, competitive moat depth, regulatory risk, and strategic positioning. A stock can score 10 based on quantitative signals while facing an SEC investigation, a CEO departure, or a patent cliff. These events are difficult to quantify before they appear in price or sentiment data.
The 60–90 day prediction window creates turnover. Monthly rebalancing based on score changes means you are trading frequently, which generates short-term capital gains tax in taxable accounts. The backtest returns are pre-tax — after-tax performance depends heavily on account type and holding period.
Crowding risk is the third limitation. As Danelfin grows in popularity, the top-rated stocks attract more capital from score-following investors. This can compress expected returns through forward-looking multiple expansion — the very signal the model tries to exploit gets priced in faster.
How to use Danelfin scores effectively
Treat the score as a screening filter, not a trading signal. Use 8+ as a minimum threshold to build a research list, then apply qualitative analysis: read the 10-K, check management compensation alignment, assess competitive positioning, and evaluate key person risk. The score tells you which stocks have the highest statistical probability of near-term outperformance. It does not tell you whether the risk-reward matches your portfolio needs.
Pair the Danelfin score with FinTara's fundamental screener for a two-layer approach: quantitative probability from Danelfin, then fundamental quality and valuation from FinTara. This combination narrows the field from thousands of stocks to a handful that pass both a machine learning probability filter and a human-interpretable value framework.