U.S. Recession Risk · point-in-time validated
A single monthly score built from eight economic indicators, computed each month using only the data that had actually been published by then. Then held to that record, without hindsight.
Example reading
Illustrative, from the validation record. The live reading is available to beta subscribers.
One number, one regime, and the reasons behind it. Published once, on a fixed schedule, and never revised after the fact.
Where this month's model output sits against its own history. Elevated from 60 to 84; High at 85 and above. Thresholds were fixed at the 60th and 85th percentiles before model selection and were not tuned to it.
An explanation panel shows which of the eight inputs pushed the score up or down: the yield curve, labor market, industrial production, housing starts, consumer sentiment and equity returns.
A separate, descriptive 0–100 reading of drawdown and volatility conditions. It says how stressed markets are now; it does not claim to predict them.
Many published recession indicators are evaluated on today's revised data, which nobody had at the time. VALTIVRA is evaluated on the figures as first released, using the Federal Reserve's ALFRED archive of historical data vintages. Four of the eight inputs (unemployment, payrolls, industrial production, housing starts) use exact historical vintages; the yield curve, consumer sentiment and equity prices are treated as unrevised.
Live publication began in September 2026. The 332-month validation record is a walk-forward simulation on archived data vintages: for each month, the model was refit on data available twelve months earlier and scored on the figures as first published. It is the same code that runs each month now, and it is the only evidence of skill the product has; we say so plainly.
AUROC measures ranking skill: the probability that a randomly chosen pre-recession month is scored higher than a randomly chosen other month. 0.5 is chance; 1.0 is perfect. The full out-of-sample series is published so the figure can be recomputed by anyone: oos_record.csv (what the columns mean).
The methodology is frozen and documented, and any change to the inputs, the model or the thresholds requires a designated new version and a full re-run of the validation.
The record includes the failures, because a track record without them is not a track record.
Professionals who already form a view on the cycle and want one disciplined, auditable input alongside their own judgement.
VALTIVRA was founded by Shane Hladinec, an independent researcher and developer focused on building transparent, data-driven tools for understanding economic and financial risk. He developed the VALTIVRA Economic Risk Engine and its point-in-time validation framework, with an emphasis on reproducibility, historical data integrity, and clearly documented methodology.
VALTIVRA was built from the ground up as an independent research product, with the model, data treatment, validation process, limitations, and historical failures documented rather than hidden behind a proprietary black box.
Independent research. Point-in-time methodology. Transparent validation.
VALTIVRA is currently a pre-incorporation project; an entity will be formed before paid launch. Shane is not a registered investment adviser and the service does not give personalised advice. An independent review of the validation record will be commissioned before paid launch and named here when engaged; qualified reviewers can request the full audit reports under NDA at shane@valtivra.org.
A small group of professional users, fewer than fifty seats, while data licensing and the independent validation review are completed. Beta access is free.
Pricing after the private beta. Professional access (single seat) — $249/month. Team (up to five seats) — $499/month. Institutional and API access — on request. Commercial pricing is expected to begin at these rates following the private beta; private beta participants will receive access under the terms of the beta program.
What we ask of beta users. Look at the score each month, tell us when it is wrong or unclear, and answer a short questionnaire at the end of the beta.