How the model is built

No. We deliberately don't use large language models for the model itself.

LLMs hallucinate confidently, are non-deterministic (same input → different output), are expensive at scale, and have no theoretical basis for predicting price movement.

Instead we use XGBoost — a gradient-boosted decision tree ensemble — on technical indicators plus a small set of macro and cross-asset features. The output is deterministic: the same input always yields the same result.

All features are computed from past-only data. There is no look-ahead leakage.

• Price-based: returns over multiple lookbacks (1d, 5d, 20d), z-scores, distance from moving averages • Volume: relative volume vs N-day median, on-balance volume • Volatility: ATR, realized volatility, Bollinger Band position • Momentum: RSI, MACD, rate-of-change • Microstructure: range/body ratios, gap features, intraday high-low spread • Calendar: day-of-week, day-of-month seasonality • Macro and cross-asset: VIX, the 10-year US Treasury yield, the US dollar index, gold and oil returns, and rolling correlation with the S&P 500 and NASDAQ

The exact set is in backend/agents/statistical.py and backend/agents/external_features.py. We do not use fundamental data, analyst targets, or news scores in the model.

Walk-forward TimeSeriesSplit cross-validation, not random K-fold. Random K-fold would shuffle the time order and inadvertently let the model train on future data — the #1 silent mistake in financial ML.

Specifically: we train on the oldest N% of data, test on the next chunk, then expand the training window forward. Every test point is data the model has never seen.

Charts and technical indicators — no forecasts, price targets or buy/sell signals.

Fails predictably on: • Earnings-day gaps — technical features can't anticipate fundamentals • Halted or suspended stocks — no recent data • First-day IPOs — no history at all • Black swan events (COVID-style shocks, war, regulatory shutdowns) • Very thin / illiquid micro-caps — noise dominates

Three reasons.

First, tabular financial features are XGBoost's home turf. The deep-learning vs gradient-boosting gap on small tabular data is well-documented: typically 1-2% in either direction, with XGBoost winning more often than not.

Second, Render's free-tier 512MB RAM ceiling won't fit TensorFlow, and we'd rather keep running costs low than pay for more.

Third, determinism and auditability. Tree models make it easier to inspect which features matter. Neural networks are harder to explain in a way a user can verify.

We use a deterministic lexicon scorer based on the Loughran-McDonald financial dictionary, plus our own multi-word phrase list. Each headline is scored ±1 and aggregated.

We explicitly do not use LLM-based sentiment. LLMs give different scores for the same headline on different calls, occasionally mark obviously negative news as positive when wrapped in ambiguous language, and are slow and expensive at scale.

A lexicon-based score isn't perfect — but it's reproducible, and we can show you exactly which words triggered the score. News sentiment scores are no longer shown in the product, and they do not feed into the model.

The models are pre-trained and pooled: the daily horizons (1d / 3d / 7d) and the long horizons (1mo / 3mo / 6mo) each load a model trained across many tickers in a market, rather than training a fresh model on every request.

On each request we:

1. Pull the latest OHLCV for the ticker 2. Compute features 3. Load the pre-trained model for that market and horizon

The pooled models are retrained offline, not on every request. The model's forward output is never shown in the product.

The model is trained to produce a point estimate and directional read for each horizon, but those forward per-stock outputs are not surfaced in the product. What you see is the technical indicators the model trains on.

No. Research only. Not financial advice. Past patterns do not predict future results. You can lose money.