Plus feature

Quantitative Models

Six quantitative models run live on any ticker, with the math visible.

Quantitative Models is the analytical engine of the platform. Load a ticker and six models run against its price history: Hidden Markov regime detection, Monte Carlo simulation, mean reversion, momentum scoring, support and resistance, and an earnings surprise predictor. Each one answers a different question and outputs in a form you can act on.

Models compute in the browser from price data - no waiting on a backend, no shared queue. Results refresh whenever the underlying data does, and the math behind each output is documented so you can decide how much weight to give the number. Same methodology applied uniformly across every ticker.

What you can do

HMM regime detection
A three-state Hidden Markov Model classifies recent price action as Bull, Bear, or Sideways with per-state probabilities and persistence statistics. Updates as new bars arrive. Useful as a regime filter - bullish strategies often perform better in identified bull regimes.
Monte Carlo + Mean Reversion
2,000-path forward simulations at 30/60/90-day horizons (Monte Carlo), and a z-score deviation gauge versus the 50- and 200-day moving averages with historical reversion rates (Mean Reversion). Two complementary views of where the price could be vs where it tends to go.
Momentum + Support/Resistance
A composite 0-100 momentum score from rate-of-change, moving-average position, volume trend, and relative strength. Plus kernel-density-estimated support and resistance levels labeled by historical touch frequency. The pair tells you where price is heading and where it'll meet friction.
Earnings Surprise Predictor
Weighted forecast of upcoming earnings beat/miss using historical surprise pattern, recent trend, consistency, and implied move from options pricing. Outputs a direction, predicted surprise magnitude, and a confidence percentage so you know how much to lean on it.

When to use it

Scenario 01
You're evaluating a long position in a stock and want a fast read on whether the market environment supports your direction.
The HMM regime indicator + momentum score together tell you if the stock is in a bullish regime with positive momentum. Two greens raise conviction; one red flags caution.
Scenario 02
A stock looks oversold to you and you want to know if it actually tends to revert.
The Mean Reversion gauge shows the z-score versus the 50- and 200-day MAs with the historical reversion rate at that deviation level. A -2σ reading with a 70% reversion rate is a different setup than -2σ at 30%.
Scenario 03
Earnings is next week and you want a model-based view on the print.
The Earnings Predictor outputs a Beat/Miss call, predicted surprise size, and confidence. Compare to the options-market implied move to see if the predicted outcome is already priced.

How it works

The HMM is fit with the Baum-Welch algorithm on the ticker's recent daily returns, with three latent states constrained to a Bull/Bear/Sideways interpretation. Transition probabilities feed the persistence metric.

Monte Carlo runs 2,000 geometric Brownian motion paths using the ticker's historical volatility as σ and historical drift as μ, projecting forward 30, 60, and 90 trading days. Percentile bands (10th / 50th / 90th) summarize the simulation distribution.

Mean Reversion computes z-scores against the 50- and 200-day SMAs, then looks back at historical deviations of similar magnitude to estimate the reversion rate within 20 trading days.

Momentum aggregates four factors: rate-of-change over 10/20/60 days, position relative to SMA 20/50, volume trend ratio, and relative strength. Each is normalized and weighted into a 0-100 composite.

Support/resistance uses kernel density estimation on historical price levels - peaks in the density indicate price levels where the stock has traded heavily and tends to react.

Earnings Predictor weights four signals: historical beat rate, recent surprise trend, consistency of surprises, and the options-implied move. The model is calibrated against the ticker's own surprise history rather than a universal benchmark.

We use cookies and similar technologies to make Ocarina Research work and to understand how it's used. You can accept all, reject non-essential, or customize. See our Privacy Policy for details.