TL;DR
AI financial ratings separated stronger and weaker stock groups, but individual returns still varied widely. In the 4,103-stock January 2026 cohort, high-rated stocks had a 7.1% median price gain versus a 12.2% median loss for the lowest band through October 12. The score describes reported business strength. Older ratings were generated retrospectively and the universe favors surviving companies; in 2025 only 32% of high-rated stocks beat SPY. Returns here exclude dividends.
- Scale
- 1 to 5, in half points
- Inputs
- SEC financial statements and earnings history
What the score measures
A NexusTrade stock rating answers one question: based on the numbers a company reported, how strong does the underlying business look?
A small, inexpensive language model reads the company's recent financial statements and writes a report that ends in a grade from 1 to 5, in half-point steps. A 4.5 describes growing revenue, healthy margins, real cash flow and manageable debt. A 1 describes a business that looks weak by those traditional measures, such as shrinking sales, steady losses or heavy debt.
Income statements
Revenue, gross profit, operating income and net income for up to 12 recent quarters, taken from SEC filings.
Growth
Three, five and ten year compound growth in revenue, earnings, free cash flow, assets and equity.
Profitability
Gross margin, net margin, return on equity and return on assets.
Balance sheet
Debt to equity, current ratio, cash and working capital.
Earnings history
Reported earnings per share against analyst estimates.
A price snapshot
Market cap and ratios such as P/E, so the report can comment on valuation. Ratings written from March to July 2025 used the price from the period they describe; reports written later use the price on the day they were written.
What a rating leaves out
The model is told to use only the financial data it is given, and it has no web search. Everything below sits outside the score.
News
Lawsuits, product launches, executive departures and earnings surprises that happen after the filings it read.
Sentiment
Social media buzz, analyst upgrades and fund flows.
The economy
Interest rates, recessions, tariffs and sector rotations.
Next year's results
A rating describes the business as reported. It makes no forecast of next year's revenue or earnings.
A rating is a dated snapshot
Every rating carries a fiscal year and the date of the newest filing it read. On the stock reports page, check both before you compare two companies. A 2023 rating says nothing about what the company reported in 2025, and a strong year can end with a weak one.
The grader is a small model, chosen because it is cheap enough to rate thousands of companies. Most stored reports come from Qwen Turbo and Gemini 2.5 Flash Lite, and newer ones from GPT-6 Luna. Small models misread numbers sometimes, so open the written report behind any score you plan to act on.
What a 1 versus a 4.5 tells you
Each stock's latest rating as of December 31, then its price change over the next calendar year. 2025: 4,561 stocks. 2026: 4,885 stocks through 2026-10-12.
Low-rated companies usually have weaker financials, and as a group their stocks did worse. In 2025, stocks rated 1 to 1.9 had a median return of -18.7%, while stocks rated 4 or higher had a median of +4.5%.
The cleanest test is this year. 4,103 ratings were written in January 2026, before almost any of 2026's returns existed. Through 2026-10-12, the median stock among them rated 1 to 1.9 is -12.2%, and the median stock rated 4 or higher is +7.1%.
How often each band lost money
Value (%)
- Rated 4 or higher 2026-10-12, 00:00:00 UTC
- 38.26%
- Rated 1 to 1.9 2026-10-12, 00:00:00 UTC
- 62.11%
The chart’s 2026 endpoint uses 4,885 stocks rated by the prior December cutoff. That is a separate cohort from the 4,103 ratings written in January 2026 discussed above.
In 2025, 61% of stocks rated 1 to 1.9 fell, compared with 44% of stocks rated 4 or higher. A larger share of the low-rated band lost money in all 12 years measured. A high rating lowered the odds of a losing year without removing them: in a broad selloff like 2022, most stocks in both bands fell.
Same rating, very different years
A rating describes a group better than it describes one stock. Pick a random 2025 stock rated 1 to 1.9 and a random one rated 4 or higher: the low-rated stock finished ahead about 36% of the time.
Only 32% of the stocks rated 4 or higher beat SPY's +16.6% that year. Some low-rated companies more than doubled on moves that last year's financial statements could not show.
Do ratings line up with future returns?
Value (%)
- Rated 4 or higher 2026-10-12, 00:00:00 UTC
- 6.43%
- Rated 1 to 1.9 2026-10-12, 00:00:00 UTC
- -13.91%
- SPY 2026-10-12, 00:00:00 UTC
- 13.9%
Read the explanation
For 2026, this chart uses the 4,885-stock December-cutoff cohort: +6.43% for the high band and -13.91% for the lowest band. The separate January-written cohort has 4,103 stocks and medians of +7.1% and -12.2%; the two cohorts should not be read as the same test.
Over the 11 completed years, 2015 through 2025, the median stock rated 4 or higher beat the median stock rated 1 to 1.9 in 10 of them. In 2020, the two medians were within a tenth of a point.
The rank correlation between rating and next-year return ranged from -0.02 to 0.27. That is a real but modest relationship. Ratings sort large groups of stocks; they leave plenty of room for any single stock to surprise.
Read the earlier years with care. Ratings for 2014 through 2023 were generated in 2025 from the filings of those years, by models whose training data may include what happened next. Most of them were priced as of the period they describe, but reports added after July 2025 carry the price from the day they were written. The universe also favors companies that still trade today. The 2025 ratings were written in March 2025, about ten weeks into that year.
The 2026 ratings written in January are the cleanest forward test, and the pattern held there: a rank correlation of 0.26 across 4,103 stocks.
Every year, by rating band
Show the yearly numbers and method
Scroll sideways to compare all columns. Select a heading to sort.
| 2015 | 2,435 | -25.5% | -10.6% | -2.7% | +1.7% | -0.8% | 0.20 | 29% |
| 2016 | 2,569 | +2.5% | +14.1% | +21.7% | +12.3% | +11.2% | 0.08 | 44% |
| 2017 | 2,658 | 0.0% | +6.3% | +12.6% | +18.4% | +18.5% | 0.15 | 36% |
| 2018 | 2,798 | -32.0% | -14.9% | -11.5% | -9.1% | -7.0% | 0.17 | 33% |
| 2019 | 2,975 | +0.1% | +12.1% | +20.6% | +25.3% | +28.7% | 0.20 | 34% |
| 2020 | 3,146 | +7.5% | +2.3% | -0.8% | +7.5% | +15.1% | -0.02 | 50% |
| 2021 | 3,362 | -7.7% | +12.8% | +20.6% | +27.3% | +28.8% | 0.19 | 32% |
| 2022 | 3,929 | -55.4% | -34.6% | -17.5% | -18.8% | -19.9% | 0.27 | 29% |
| 2023 | 4,113 | -17.0% | +2.0% | +4.9% | +12.7% | +24.8% | 0.18 | 31% |
| 2024 | 4,296 | -18.8% | -2.8% | +6.3% | +9.1% | +24.0% | 0.19 | 34% |
| 2025 | 4,561 | -18.7% | -4.2% | +0.9% | +4.5% | +16.6% | 0.15 | 36% |
| 2026 (to 2026-10-12) | 4,885 | -13.9% | +3.0% | +1.8% | +6.4% | +13.9% | 0.25 | 33% |
No matching rows. Clear the filter to see all records.
Rating: each stock's latest report dated on or before December 31 of the prior year, with a fiscal year no older than three years. Return: split-adjusted price change from the first close in the first two weeks of January to the last close of the year, dividends excluded. Stocks without an early-January close are excluded; stocks that stopped trading keep their return to the last close. 'Low beat high' is the share of random pairs in which a stock rated 1 to 1.9 outperformed one rated 4 or higher.
How to use the ratings
Use them to narrow a list
Sort thousands of stocks down to the companies with stronger reported financials, then do the rest of your research on that shortlist.
Read the report behind the number
The written analysis shows which numbers drove the grade and where the model may have misread something.
Check the date and fiscal year
Compare ratings from the same fiscal year, and look for filings or news that came after the rating.
Add what the rating leaves out
Valuation, recent news, the economy and your own time horizon all matter more over a few months than last year's statements.
Test a rule before you trade it
Backtest a portfolio that buys high-rated stocks and compare it with a simple index fund over several years.
Questions
Is a high rating a buy signal?
No single rating is a trade. It says the reported business looks strong. In 2025, 44% of stocks rated 4 or higher still lost money.
Why did a stock rated 1 go up?
Short-term prices move on news, sentiment and expectations. A weak business can rally on a turnaround story or speculation, and the rating only reads what was already reported.
What does a rating of 3 mean?
It describes a middling business on the reported numbers, with some strengths and some weaknesses. In most years the 3 to 3.9 band landed between the high and low bands.
Does the rating include valuation?
It does, in part. The report sees a price snapshot with P/E and market cap and can comment on them, but the grade is mainly about the business.
Which AI model writes the ratings?
Small, low-cost models: mostly Qwen Turbo and Gemini 2.5 Flash Lite for existing reports, and GPT-6 Luna for new ones.
How many stocks are rated?
5,058 US stocks had a rating dated on or before December 31, 2025.
No comments yet.