Disclosures

Opulence Alpha provides quantitative research and software tools. Nothing on this site is investment advice, a recommendation, or an offer to buy or sell any security.

Forecasts and fair values are model estimates. The studies are measured tendencies across many stocks, published with their limits. The platform does not decide for you: decision and execution authority is always yours.

Opulence Alpha is non-custodial research software. It does not hold client money or assets, does not connect to your broker and does not place orders.

The Golden Tickets and the Growth, Moderate and Conservative Books are the platform's own research, and they hold no client money. A test portfolio an account follows holds no money either. Acting on any of it is your decision, placed through your own broker.

The Golden Tickets are the platform's weekly benchmark engine. They measure what the forecasts alone are worth, before any portfolio is built around them.

After each Monday's close, the 1-, 5-, 21- and 63-day forecasts are blended into one conviction score at 22.4%, 15.8%, 24.8% and 37.1%, weights the platform's search engine found rather than a person set. The top ten names by that score form the basket, and a name with a negative forecast is held short. The basket is held for the trading week and followed exactly as picked.

No costs are charged: no commission, slippage or borrow. Results are gross of costs, in percent of the basket, and compared with the S&P 500 as price return (SPY), so index dividends are left out. Every closed week, whether it rose or fell, is published on the Refutation Ledger at /ledger with every name, side, entry and exit price.

Growth, Moderate and Conservative are the platform's three model portfolios, one for each risk profile the sign-up questionnaire can assign. They are run every trading day and hold no client money. A strategy is validated at the Promotion Gate before any Book may express it.

Their figures are net of modelled trading costs, charged on every fill: each stock's own measured bid-ask spread, capped at 60 basis points; a 0.5-basis-point commission; and a market-impact charge on orders large enough to move the price.

Each Book's value is published every trading day beside the S&P 500 and the Nasdaq 100, both as price return (SPY and QQQ), so index dividends are left out. How the Books are built is set out on /books.

The studies are historical measurements on one universe: 1,767 US-listed common stocks across 11 sectors, observed every Wednesday since January 1995. Companies that later delisted stay in with their history, to limit survivorship bias.

A study describes a tendency across hundreds of stocks. None forecasts a single stock's price, and even a proven relation is only a nudge for any one stock.

Results are before trading costs, taxes and market impact. Signals are recomputed over the past with today's definitions, so a relation found is a measured tendency, not the record of a strategy run live. The full list of limits is on /studies/methodology.

Fair value is a model estimate, recomputed after every US trading day. For most companies it is a discounted cash-flow (DCF) valuation: cash flows projected from the latest annual report and discounted at a CAPM-based cost of capital, blended with the company's own earnings and EBITDA multiples and its book value.

The estimate carries its uncertainty. Each company is run through 10,000 scenarios, each with its own growth and discount rates, and the range from the 5th to the 95th percentile is recorded around the central value.

It is a long-run value estimate: not a forecast of where the price will trade on any date, not a price target and not a recommendation to buy or sell. Names whose cash-flow value fell below the three-month model's price range, and names the model flags as implausible, are left out of the figures on /fair-value, so those figures exclude the most bearish readings.

The Regime Radar reads nine axes of the US market every trading day and fuses them into one composite regime. Its forecasts are of the regime, never of a price: the chance the composite changes within five sessions, with the model's interval.

The early-warning level is a weighted blend of four macro warning signals. It is a level, not a probability.

The sector tables show how US sectors did against the S&P 500 in past sessions with the same regime, on next-session ETF price returns. They describe history, not what will happen.

Forecasts come from predictive machine-learning models trained on market, filing and macroeconomic data: gradient-boosted decision trees (CatBoost), LSTM neural networks (PyTorch) and XGBoost quantile models whose ranges are calibrated by conformal quantile regression. Bernstein Online Aggregation combines their forecasts. Regimes come from a percentile classifier on each axis, fused by a hidden semi-Markov model.

No generative AI is used in the forecasts. No language model writes, summarises or decides anything in the research, and the platform calls no ChatGPT, Gemini or Claude service. The one language model in the system is FinBERT, a classifier that scores the tone of SEC filings as one input among many. It writes nothing.

Forecasts are estimates, not certainties. They are published with a range wherever a calibrated interval exists, and scored against what happened once they mature. A model that fails its statistical gate on held-out data carries no trading weight.

The research runs on daily closing prices of US-listed stocks, adjusted for splits and, where the source records them, dividends. Fair value and the Regime Radar are recomputed after each US trading day, the Books are run every trading day, and each of those pages states the date or the close it reflects.

Company fundamentals are read as of their SEC filing dates, not the period they describe. Macroeconomic series come from the St. Louis Fed's FRED database and enter on the day they were published, so no reading uses data that was not yet out.

Coverage is US-listed common stocks on the NYSE and Nasdaq. The Turkish pages present the same US research in Turkish; they do not cover Turkish companies.

The figures on this site are measured over stated windows, and the window is given with them. Past performance does not guarantee future results, and all investing involves risk of loss.

By the platform's own standard, a live record of a few months shows that the system operates as designed. It does not establish a statistically significant edge, and the site does not present it as one.

Questions about anything on this page go through the contact form at /contact, and a person reads every message. How to report a security issue is set out in /.well-known/security.txt.