How It Works

The first gate is entirely quantitative.

The PTP universe. PTP does not scan every listed stock. A stock or ETF enters the universe of roughly 300 names only if it has weekly listed options, a share price above $20, and at least ten years of daily price history to build its models from. Those requirements leave out penny stocks and thinly traded names that lack an active weekly options market, and keep the focus on liquid, well-established securities with enough history to model.

For every stock in the PTP universe, our models ask:

What is the probability that this stock will close higher five trading days from now?

To pass Gate 1, the model probability must be 0.75 or higher.

A probability of 78% does not mean the stock is expected to gain 78%. It means the model estimates a 78% probability of a higher close at the end of the five-trading-day forecast window.

Stocks do not share one universal model. Each stock has models developed from its own historical market behavior.

The result of this stage is the Full ML Table.

A high statistical probability is the starting point, not the finished selection.

Stocks passing Gate 1 are next evaluated against separate fundamental and analyst criteria, including:

• earnings-estimate momentum,
• analyst consensus, and
• proximity to the next earnings report.

Stocks reporting earnings within approximately one week are excluded from new selections.

The fundamental information does not alter the machine-learning probability. A stock with an 82% ML probability remains an 82% ML probability whether or not it passes Gate 2.

Instead, Gate 2 asks a different question:

Is the current fundamental and analyst picture consistent with allowing this high-probability statistical setup to move forward?

Stocks passing Gates 1 and 2 form the ML + Fundamentals list.

ETFs do not have company earnings or conventional company fundamentals. When those measures do not apply, the company-specific portions of Gate 2 are not used. ETFs continue through the applicable quantitative, current-event and human-review stages.

Historical data has an unavoidable limitation:

It cannot know what happened last night.

A company may announce an acquisition. An analyst may change a rating. A government may announce a policy change. Commodity prices may move sharply. A geopolitical event may alter an industry’s outlook.

The historical ML model cannot see information that occurred after its data cutoff.

Gate 3 is designed to address that limitation.

Two AI reviewers are each given the quantitative setup, relevant fundamental and analyst information, and current public information. Each evaluates the complete situation separately, including recent news, company developments, scheduled events, sector conditions and broader developments that may matter to the stock.

Unlike the machine-learning model, the AI reviews are deliberately sighted.

Their purpose is not to rediscover the ML probability. Each reviewer judges whether information available now supports a higher price over the next 30 days, or contradicts or materially complicates the quantitative case. A stock moves on only when both reviewers agree that current information supports it.

AI does not calculate or change the ML probability.

The probability remains the output of the machine-learning model.

The AI reviews add information the historical model could not have known.

Every stock reaching the final gate receives a human review by Dr. James Krider before publication as a Four-Gate selection.

The complete picture is considered: the machine-learning probability, fundamental and analyst information, current-event review, and any unusual circumstances that warrant additional attention.

This is not an attempt to replace the model with intuition.

It is a final quality-control step before publication.

Stocks that clear the complete process become the Four-Gate Selections.

One Four-Gate selection is featured each trading day as the free Daily Pick.

A probability describes uncertainty. It does not eliminate it.

Suppose today’s model probability is 75%.

That does not mean the stock will rise 75%. It does not mean PTP knows that today’s stock will finish higher. And it does not mean that every group of four 75% forecasts will contain exactly three winners.

It means the model currently estimates a 75% probability of a higher close five trading days later.

There is an important distinction between two numbers shown by PTP:

For example, a stock may receive a model probability falling within the 80%–85% probability band. The historical statistics displayed for that band show how past forecasts in that range actually performed.

The two numbers should not be confused.

The model probability is the forecast. The historical win rate is evidence about how forecasts in that probability range have behaved.

No probability eliminates losing outcomes. Market conditions also change. During unusual regimes—wars, financial crises, abrupt policy changes or other major disruptions—future results can differ substantially from historical experience.

Each Four-Gate selection includes a 90% CSP/Wheel level for the applicable Friday expiration.

This is a separate calculation from the 5-day ML probability.

For each stock, PTP examines its own historical price behavior and estimates a discount below the morning opening price associated with the stock having closed above that level on the applicable expiration date in roughly nine of ten comparable historical windows.

That sentence is just as important as the 90%.

The level describes historical behavior. It is not a guarantee, and unusual market conditions can produce outcomes substantially different from the historical record.

The level also does not mean the stock remained above that price throughout the holding period. The measurement is based on where the stock closed on the applicable expiration date.

PTP separates the process so you can see what happens at each stage.

Except for the ML probability, these statistics describe all past forecasts in the same probability band, not the individual stock’s own trading history.

Probability Trader Pro is built around probabilities, not certainty.

Each layer has a defined job.

Models are tested only on data that comes after their training data.

The model stays blind to the future.

The final review does not stay blind to information.

PTP’s methodology is intended to be inspectable.

For readers who want to go beyond the summary on this page, our research library provides additional information about the models, probability methodology, historical testing and the use of probabilities in trading research.

The methodology is easier to understand when you can see the numbers.

Create a free account to see the Daily Pick, or join PTP Pro for access to the full probability research, filtered selections and CSP/Wheel levels.