AI-modelled market context, footnoted like research^N1Note N1 · MethodEvery published statement carries a footnote: model class, training cutoff, input data window, and the empirical reference. Hovering surfaces the citation, the way a research paper would. No black boxes.
never as advice^N2Note N2 · Legal postureLumequant S.r.l. is not authorised by Consob and is not registered under MiFID II as an investment firm. Nothing on this site or in the daily brief constitutes personalised investment advice. Decisions are yours..

No. Nothing published on lumequant.com, in the daily brief, in the signal cards, or in regime-break notes is investment advice. Lumequant S.r.l. is not authorised by Consob and is not registered under MiFID II as an investment firm.

What we publish qualifies as analisi finanziaria di carattere generale under Italian law — general financial analysis addressed to the public. We do not know your portfolio, your risk tolerance, your tax position, or your time horizon. We do not adapt anything to you personally. You decide.

Equities: the major global indices (SPX, NDX, RUT, SX5E, FTSE, NKY, HSI, KOSPI) plus the top 200 by market cap in each region. Daily.

FX: all G10 majors and the most liquid EM crosses (USD, EUR, JPY, GBP, AUD, CAD, CHF, NZD; plus MXN, BRL, ZAR, INR, TRY, KRW, CNY). Real-time during European/US hours.

Crypto: top 20 by market cap, including funding rates and CEX/DEX flow proxies. Real-time, 24/7.

Commodities: gold, silver, oil (WTI + Brent), copper, natural gas. Daily.

Rates: US Treasuries (2y/5y/10y/30y), Bund + BTP curves, JGB 10y. Daily.

Each signal card identifies its asset universe precisely.

Signals come from a stack of model classes appropriate to the signal type: gradient-boosted regime classifiers (for cross-asset momentum / mean-reversion), Bayesian state-space models (for volatility regimes), and NLP-extracted text features (for macro and policy overlays). Every signal card discloses the model class, training cutoff, out-of-sample window, and cross-validation protocol.

We do not publish backtested return statistics. They are systematically misleading: they invite look-ahead bias, survivorship bias, and parameter-overfitting bias — and they prime readers to trust models more than the conviction bands warrant. Instead, every signal is published with an explicit "what could break it" section listing thresholds we are monitoring in real-time.

Past performance is not indicative of future returns. The conviction band describes the model's belief about its own next-window accuracy under the regime classifier; it is not a probability of profit.

Lumequant S.r.l. is a Milan-incorporated società a responsabilità limitata, P.IVA + REA Milano (see footer). The supervisory authority for data protection is the Garante per la protezione dei dati personali in Rome.

The data we collect from readers is the minimum to deliver the brief: name, email, and (if you upgrade) billing. We do not sell data, do not share it with brokers, do not use it for ML training, and do not enrich it with third-party profiles. Account data is retained per Codice Civile art. 2220 (10 years for accounting records); marketing data is purged on unsubscribe or after 24 months of inactivity, whichever is sooner. Full posture in the Privacy modal in the footer.

No. Lumequant is a publication. It does not connect to broker accounts, does not place trades, does not manage portfolios, and has no plans to. The closest thing to "action" we offer is an email link to the relevant signal card; the route from a signal to a trade — including whether to take one at all — is entirely yours.

This is a structural choice, not a missing feature. It keeps us outside MiFID II investment-firm scope, keeps the regulatory surface narrow, and keeps the relationship clean: we provide research, you make decisions.