How It Works

The Curistat Volatility Rating is our flagship metric. Every trading session gets a rating from 1 (dead calm) to 10 (buckle up) based on 57 curated features extracted from market data.

We started with a much larger feature set and ruthlessly pruned it down to only what moves the needle. No kitchen-sink modeling here. Every feature earns its place or gets cut.

The final rating maps to one of five session-type buckets, each with its own expected range distribution. Low ratings mean tight ranges. High ratings mean wide ranges and real opportunity (or real risk, depending on how you trade).

We show our results, not our recipe. The accuracy metrics, validation methodology, and historical track record are all public. The exact feature weights and model internals stay in-house. That is our edge, and we protect it.

10 years of 1-minute OHLCV data for ES (E-mini S&P 500) and NQ (E-mini Nasdaq 100) futures. Millions of bars per product, covering 2016 through 2026. We use institutional-grade data from DataBento.

Full VIX family data (VIX, VVIX, VIX9D, VIX3M, SKEW) going back to early 2015. That gives us thousands of trading days of continuous data across every volatility regime you can think of.

Economic event data covering 30+ event types (CPI, FOMC, NFP, GDP, PMI, and more) with actual vs. consensus values so we can measure surprise impact.

Treasury auction data for major maturities including bid-to-cover ratios and tail measurements. If it moves the market, we track it.

All data updates automatically every day. We run staleness checks and integrity validation so stale data never sneaks into a forecast.

Our model achieves a test correlation of r=0.84. That means it explains roughly 70% of the variance in actual session volatility. Not perfect, but substantially better than guessing or vibes.

We validate using rolling cross-validation across dozens of windows. Every single window cleared the bar, including through COVID (2020), the 2022 bear market, and the 2023-2024 recovery. This is not a model that only works in hindsight or in one regime.

Volatility clustering features dominate the model, which confirms what experienced traders already know: yesterday matters. Recent volatility is the single best predictor of what comes next.

Our directional signals have documented win rates between 65% and 88%, each grounded in academic research and validated on our own data.

Every day, we compare what we predicted against what actually happened. No exceptions, no skipped days, no convenient omissions.

A prediction counts as a hit if the actual session volatility lands in the predicted bucket or one bucket away. For expected ranges, we check whether the actual range fell inside our confidence interval.

We publish a 30-day rolling accuracy log. Wins and misses, all of it. If we had a bad week, you will see it.

Our accuracy methodology is documented and consistent. We do not retroactively adjust predictions or drop inconvenient data points. That is not how you build trust.

We do not give financial advice. Curistat is an analytics platform. Your trading decisions are yours.

We do not predict direction with certainty. Our directional signals show statistical edges. A 72% win rate still means you are wrong 28% of the time.

We do not hand you trade signals with entry and exit prices. We give you context: expected volatility, regime, events. You make the call.

We do not guarantee anything. Markets evolve, regimes shift, and edges decay. We monitor and adapt constantly, but no model is bulletproof. Anyone who tells you otherwise is selling something.