AI trading signals

A Cashora signal is a probability score, not a buy order. For a given ticker it says which direction the recent price structure leans, how strongly, and — most importantly — whether that kind of call has actually worked on that ticker before. Nothing is executed: Cashora is a market probability signal predictor, not a broker.

• Trend — is price holding above or below its longer-run average?
• Momentum — is the pace of movement building or fading?
• Overbought / oversold — has the move stretched far enough to snap back?
• Breakouts — has price pushed outside its recent normal range?
• Average crossings — has a faster average crossed a slower one?

Each method votes buy, sell or hold with its own strength. The votes are combined into a single consensus call and a confidence figure. Every page that shows a signal — Predictions, a stock's own page, and Today's Top Picks on the home page — runs that same engine on the same daily price history, so the numbers agree everywhere.

A raw score only describes today's chart. Before it is shown, Cashora replays the same engine through five years of that ticker's history on a 20-day holding period and asks three questions: did scores in this band actually win, does this ticker pay in this direction at all, and is the price above its 200-day average? If the answer is no, the score is reduced and labelled — never reversed. An unreliable signal is reported as low confidence, not traded backwards.

Each signal carries a grade:

• A — measured history supports this call.
• B — modest measured support.
• C — unproven on this ticker.
• D — history shows no edge here; information only.

Measured across 30 US stocks and ETFs over ten years of daily data with a 20-trading-day holding period, actionable signals were directionally correct 58.8% of the time over 2,289 positions, averaging +1.52% per position at a mean Sharpe ratio of 0.59. On the 15 tickers held back from every tuning decision, the hit rate was 57.3% at a mean Sharpe of 0.38, with 12 of 15 profitable. Without the directional, trend and confidence filters the same engine measured roughly 51% — essentially noise. Short calls measured about 45% and are therefore never marked actionable.

The full per-ticker table, including the names where the model loses money, is published on the measured performance page.

Where it works and where it does not is not uniform. Volatile, range-heavy names have shown the clearest edge; strong long-term compounders have repeatedly underperformed simply holding the stock. Trend and crossing methods fire almost daily and sit near a coin flip; the overbought/oversold and breakout methods fire rarely and carry most of the real signal. You can reproduce all of this yourself on the Backtesting page.

• Not financial, investment, legal or tax advice.
• Not a broker — no order is ever placed, and no account is ever connected.
• Not a forecast. Hypothetical and backtested results do not predict future returns.
• Not a guarantee. Every measured figure above comes with real losing periods and drawdowns.

Full detail is in our Terms of Service.

How often do signals update? Once new daily closing data arrives, so a signal reflects completed sessions rather than intraday noise.

Which stocks are covered? Every US-listed ticker, not a curated shortlist — browse them in Markets or the Gem Finder.

Can I see the chart behind a signal? Yes — the Signal Terminal shows the price history and indicators the score was built from.