How a signal is measured

Every figure beside a name on Trade Cards comes from one method. The models rank the names, and a record says how often names ranked like that finished on their side. This page walks through it in the order a trader asks.

At a glance

• The side. Bullish is the top of tonight's ranking, bearish the bottom. They are the desk's words for the buy side and the sell side.
• The signal. A name in the strongest 5 percent at 45 trading days, or the strongest 2 percent at 20. Realized volatility must also be over 25 percent a year.
• The probability. How often names in that band finished the horizon on their side, on trading days the model never trained on. Any stock sits beside it.
• Confidence. A word for the band, not a second probability. High is the strongest 1 percent, Medium the 2 or 5 percent band.
• Why near 60. Any stock rose over 45 trading days 56 times in 100. The strongest bullish band rose 60 times in 100, 4 points above that.
• Standard or ours. The testing method is standard research practice. The models, the cut points and the words are Ohey's own.
• Index funds. A fund prints a probability only from its own record. A short record reads Not measured, and a record with no edge reads Neutral.

Key difference: the probability is a measured share of past names ranked like this one. Confidence names the band that share comes from.

How buy and sell are decided

The side comes from the ranking alone. Bullish is the desk's buy side and bearish its sell side. Every evening the models score about a thousand US stocks and rank them against each other at 45 and at 20 trading days. The top of the ranking is bullish, and the bottom is bearish. A rank from about 20 to 80 sits too near the middle to name a side.

Nothing else votes on the side. The chart, the option chain and the news are context on other panes, and none of them changes it. Trade Cards prints the side on every row.

A rank is a place in that evening's queue, not a price forecast. A name sits where it does because of how it compared with the other names that evening.

On Trade Cards, a bullish card carries a structure that gains when the shares rise. A bearish card carries one that gains when they fall. Each card states the exact structure, limit and exit the record measures.

When a rank becomes a signal

A signal needs two things.

• A strong rank. The strongest 1, 2 or 5 percent at 45 trading days, or the strongest 1 or 2 percent at 20. The 10 day rank failed its own holdout test, so it never makes a signal.
• Enough movement. Realized volatility over 25 percent a year. A quiet stock at the top of the ranking reads Low RV, not a signal.

A name short of a signal still carries an outlook. That is its 5 percent slice of the ranking, with the slice's own measured record.

What the probability is

The probability is a count from the record. It covers every stock day the model scored on trading days it never trained on. That record runs from November 2022 to July 2026 and holds about a million stock days. For each band, it counts how often names finished 45 trading days later on their side.

Any stock sits beside it. That is how often every scored name did the same, whatever its rank. The gap between the two is the edge.

At 45 trading days the record reads:

At 20 trading days only the strongest 1 and 2 percent are offered. They rose 58 times in 100 against 54 for any stock, and fell 51 and 50 times against 46.

What Confidence means

Confidence is a word for where the figure comes from. It is not a second probability, and it is not how sure the model feels.

• High. A signal in the strongest 1 percent.
• Medium. A signal in the strongest 2 or 5 percent.
• Low. An outlook whose slice sat 2 points or more above any stock, and held that side in 4 of 5 years.
• No edge. A slice that did no better than any stock. The row reads Neutral.

The models also emit a confidence number of their own. It barely moves across a thousand names, so the desk never prints it.

Key difference: Confidence says how strong the evidence behind a row is. The probability states what that evidence measured.

Why the probability stays near 60

The number cannot be read on its own. Stocks drift up, so any stock rose over 45 trading days 56 times in 100. The strongest bullish band rose 60 times, 4 points above that. The strongest bearish band fell 57 times, where any stock fell 44: 13 points above.

Whether one stock is higher or lower in nine weeks is mostly noise, even for the best models. A steady few points above any stock is a real edge in quant research. Here it holds across a thousand names an evening and nearly four years of trading days the model never saw.

A much higher figure on stock direction would be the warning sign. It usually means the test saw the answers, or the rule was fitted to one stretch of market.

The higher figures on a card measure something else. Probability of profit on a credit spread, often 70 or more, comes from the option market's own prices. It is the chance the market prices that the shares stay past the short strike, not the model's figure.

Standard method or Ohey's own

The testing is standard research practice. The models and the cut points are Ohey's own.

The standard parts:

• Ranking names against each other each evening. The usual way quant equity research reads a model, rather than a price forecast for each name.
• Testing only on unseen trading days. The record walks forward in time. Each test period sits after its training period, with about a year between them, so nothing leaks.
• A measured share beside the base rate. How often each band was right, next to how often any name was. Weather forecasters publish the same kind of table.
• A check across years. An outlook's edge must hold in 4 of 5 calendar years, not just on average.
• Probability of profit from delta. The figure most brokers print for an option trade.

Ohey's own parts:

• The models. Ohey's own machine learning models and their inputs, trained on Ohey's own price and options history.
• The cut points. The 1, 2 and 5 percent bands, the 25 percent volatility floor, the 2 point edge, 4 of 5 years, and 10 stretches for a fund. Each was fixed before any figure was read.
• The words. High, Medium and Low, and the seven states such as Signal, Outlook and Neutral.

Why an index fund can read Not measured or Neutral

SPY, QQQ, IWM and DIA are ranked beside the stocks at 45 trading days. The top 20 percent reads bullish, and the bottom 20 percent bearish.

A fund never borrows a stock's probability. Funds rise far more often than single stocks, so a stock's figure would mislead. DIA rose over 45 trading days 74 times in 100 on its own record, where any stock rose 56. So a fund prints a probability only from its own record. It needs at least 10 separate stretches of 45 trading days.

• Not measured. SPY, QQQ and IWM joined the model's data in August 2025. Their own records start in December 2025 and hold 4 stretches, too few to state a figure. It is not missing options data: every index row still carries its put credit spread.
• Why the records are short. The model is being retrained on their full price history, which will lengthen their records.
• Neutral. DIA's bullish side is measured, with 15 stretches since November 2022. It rose 62 times in 100, against 74 on any day of the same record. That sits below its own any day, not below any stock, so DIA reads Neutral. Its row on the Indexes tab of Trade Cards says so.
• The rule. A fund reads Neutral whenever its record sits less than 2 points above its own any day. It is the same rule a stock slice with no edge follows.

Theme slots with no stock

Themes list ten watch lists by line of business, each in named categories. A category with no listed stock is left off the screen. AI Compute Marketplaces is one. The companies that run GPU compute marketplaces are private, or trade only as crypto tokens, so no listed stock or option fits.

Frequently asked questions

How does the desk decide buy or sell? From the ranking alone. Each evening the models rank about a thousand stocks at 45 and 20 trading days. The top is bullish, the buy side, and the bottom bearish, the sell side. Trade Cards prints the side on every row.

Is a Bullish signal a buy recommendation? No. The desk publishes the same research for every reader, never personal advice. A signal states where a name ranks and what names ranked like it did next.

Why is the probability lower on the bearish side? The probability beside a bearish row is how often names ranked like it fell. It is not a probability of profit. Stocks rise more often than they fall, so any stock fell over 45 trading days only 44 times in 100. The strongest bearish band fell 57 times. That is a smaller figure than the bullish side's, but a larger edge.

Why does Confidence read Low on a row with a probability? That row is an outlook, not a signal. Its slice sat 2 points or more above any stock in 4 of 5 years. It missed the signal rule on rank or on volatility.

Why does the card say 75 in 100 when the signal says 59? They measure different things. Probability of profit on a credit spread is the option market's own price for the shares staying past the short strike. The 59 is how often names ranked like this finished on their side.

Related panes

• Trade Cards prints the side, the probability and the Confidence word on every row.
• Words the desk uses defines every state and word in one place.
• Rank Not Return explains what a rank is.
• Checking the record shows how the paper record is kept.