AI-native risk engine
Institutional-grade risk math — VaR and CVaR, factor decomposition, options Greeks, stress testing, credit and liquidity risk — computed from live market data and interrogable in plain English. Built for RIAs and boutique managers priced out of the enterprise risk platforms.
Ask it like an analyst
Ask Quants runs your question against the full engine — the same VaR, factor, options, and credit models behind every workspace — and answers with the computed numbers, not summaries of someone else's research.
How much of my portfolio's risk is just the market?
Factor decomposition — Fama-French three-factor regression splits your volatility into market, size, and value exposure versus true idiosyncratic risk.
What happens to this book if 2008 repeats?
Historical stress testing — replays named crisis windows through your actual positions, alongside VaR and CVaR computed four independent ways.
Could this company's balance sheet survive a downturn?
Structural credit risk — a Merton distance-to-default model infers default risk from equity behavior and reported debt, no bond or CDS data required.
Platform
VaR and CVaR by four methods — historical, parametric-normal, Student-t, and Monte Carlo GARCH — with rolling out-of-sample backtesting (Kupiec proportion-of-failures) so you know which method to trust at which confidence level.
Fama-French three-factor decomposition: how much of a position's risk is market beta, size, value — and how much is genuinely its own.
Full Greeks and implied-volatility surfaces — term structure and put skew across expiries — built on real-time options data, with degenerate same-day expiries filtered out rather than averaged in.
Named historical crisis scenarios replayed through current positions, plus worst-single-day analysis over any window.
Days-to-liquidate at 5/10/20% of average daily volume, Amihud illiquidity ranked against the security's own history.
Merton/KMV structural model — implied default probability and distance-to-default from equity data and reported debt.
HHI and effective-N concentration measures, single-name limit checks, and empirical FX beta against major dollar pairs.
Real-time market and options data from Polygon on paid institutional tiers, behind a multi-provider fallback chain with circuit breakers — a health check runs before every demo we give.
Validation
Every risk model in Quants is validated against independent reference calculations — VaR, CVaR, Greeks, and factor loadings reproduce independent implementations to within 1×10−6 — and the backtests run strictly out-of-sample, with no lookahead.
We publish what fails. Our own backtesting found that parametric-normal VaR understates 99%-confidence risk — it breached at twice its claimed rate and failed the Kupiec test — while historical VaR stayed calibrated at both levels. The engine tells you that, because the answer to “which VaR should I quote?” is worth more than a dashboard that never disagrees with itself.
The same discipline applies to nulls: when Apple's dollar sensitivity came back statistically insignificant, the engine reported no significant exposure rather than manufacturing a beta to fill a cell.
| Method | Conf. | Breach rate | Expected | p-value | Verdict |
|---|---|---|---|---|---|
| Historical | 95% | 6.99% | 5.00% | 0.054 | PASS |
| Parametric-Normal | 95% | 6.39% | 5.00% | 0.171 | PASS |
| Historical | 99% | 1.80% | 1.00% | 0.107 | PASS |
| Parametric-Normal | 99% | 2.00% | 1.00% | 0.049 | FAIL |
Quants also ships exploratory ML price-prediction and backtesting tools. Under the same honest, lookahead-free testing, they did not show a reliable predictive edge — so they are labeled exploratory in the product and never sold as alpha. A risk engine you can trust starts with a vendor that doesn't grade its own homework generously.
Design partner program
Quants is in its pilot phase. We're looking for a handful of RIAs and boutique managers to use it on real work and shape what gets built next.
8–12 weeks of complete access — every risk workspace, market data, and the Ask Quants analyst. No cost, no card, no long-term commitment.
A 30-minute call every two weeks is the entire cost. What design partners need built is what gets built — this is a partnership, not a request queue.
You get the validation report, the known limitations, and straight answers. We ask for the same: “I stopped using it because X” is the most valuable thing you can tell us.