# How to Evaluate the Quality of Stock Market Recommendations

> One framework that works for Wall Street analyst ratings, newsletters, AI stock ratings and tips on social media: seven tests, a comparison of the sources, and a checklist to run before you act on any call.

Source: https://www.exent.ai/resources/how-to-evaluate-stock-recommendations · Reviewed by David, Senior Analyst, Exent · Published Oct 27, 2026 · Updated Oct 27, 2026

## Key takeaways

- Judge any source on seven tests: hit rate against a benchmark, average return above that benchmark, number of calls, a fixed time horizon, every call counted, conflicts disclosed, and results after costs.
- A hit rate needs a base rate. In Exent’s 7-year backtest of 988,127 analyst ratings, 64.4% of rated stocks were higher 12 months later, but in a mostly rising market most stocks go up anyway.
- Sample size decides whether a record means anything. Illustrative: with 50 calls, a 60% hit rate is statistically hard to tell apart from anything between about 46% and 74%.
- Most sources fail at least one test: 79% of active large-cap US funds trailed the S&P 500 in 2025 (SPIVA), and a study of stock-tip posters found 56% had negative skill.
- Agreement among several sources with proven, complete records is a stronger signal than any single call.

## How do you evaluate the quality of stock market recommendations?

Ask for the full record and run the same seven tests on it, whoever made the call. A recommendation is a prediction, so its quality is measured the way any prediction is: how often it was right, how much it earned when right and wrong, over how many tries, against what benchmark, and whether anything was left out or paid for.

| Test | Question to ask | What good looks like |
| --- | --- | --- |
| Hit rate vs a benchmark | How often did the pick beat the index, not just rise? | Clearly above the base rate for the same period |
| Average excess return | What did the average call earn above the index? | Positive across all calls, not a few big winners |
| Sample size | How many calls stand behind the record? | Hundreds, over several years and market moods |
| Time horizon | Was the horizon stated when the call was made? | A fixed horizon, judged on that date |
| Completeness | Are losing and deleted calls included? | Every time-stamped call, with nothing removed |
| Conflicts | Is the source paid, or holding the stock? | Disclosed in writing, with the amount |
| After costs | What was left after fees, spreads and taxes? | Still ahead of a cheap index fund |

## How do analyst ratings, newsletters, AI ratings and social tips compare?

They differ mostly in how easy they are to check. Wall Street ratings are public and dated, so their records can be rebuilt call by call. Social-media tips are the hardest: posts get deleted and horizons are rarely stated.

| Source | Can you check the record? | Usual horizon | Main conflict | What the evidence says |
| --- | --- | --- | --- | --- |
| Wall Street analyst ratings | Yes: ratings and targets are public and dated | About 12 months | Investment banking ties, access to management | Top consensus picks earned over 4% a year before costs in a classic study, but not reliably more than zero after costs |
| Active fund managers | Yes: audited returns | Years | Fees | 79% of active large-cap US funds trailed the S&P 500 in 2025 |
| Stock-picking newsletters | Partly: usually self-reported | Varies | Subscription marketing | Studies of an independent archive found no edge for the group |
| AI and quant stock ratings | Partly: mostly vendor backtests | Often weeks to months | Model marketing | Backtests are hypothetical; AI claims themselves need checking |
| Social media and forum tips | Rarely | Usually none stated | Undisclosed positions or paid promotion | In one study, 56% of posters had negative skill |

Sources: [Barber, Lehavy, McNichols and Trueman (2001)](https://ideas.repec.org/a/bla/jfinan/v56y2001i2p531-563.html) on consensus ratings, the [SPIVA U.S. Year‑End 2025 scorecard](https://www.spglobal.com/spdji/en/spiva/article/spiva-us-year-end-2025/) and the [“Finfluencers” working paper](https://www.sfi.ch/en/publications/n-23-30-finfluencers) (Swiss Finance Institute, 2023). The newsletter research is summarized in do newsletters beat the market, and AI scores in [analyst ratings vs AI stock ratings](https://www.exent.ai/resources/analyst-vs-ai-stock-ratings).

## What is a good hit rate for stock picks?

One that clearly beats the base rate for the same period and the same kind of stock. “The stock went up” is a low bar when the market rises in most years. The table puts common hit-rate measures next to what an unskilled pick would have scored.

| Measure | Result | Source |
| --- | --- | --- |
| Analyst-rated stocks higher 12 months later | 64.4% | Analyst backtest, 7 years, to Nov 16, 2026 |
| Analyst price targets reached within 12 months | 53.0% | Analyst backtest, 7 years, to Nov 16, 2026 |
| Today’s largest US stocks that rose over 12 months | 345 of 475 | Market data, 12 months to Oct 30, 2026 |
| Today’s largest US stocks that beat the S&P 500 over 12 months | 262 of 475 (55.2%) | Market data, 12 months to Oct 30, 2026 |
| Analyst targets met at the end of 12 months / at any point within them | 38% / 64% | Bradshaw, Brown and Huang (2013), targets from 2000–2009 |

The backtest and market-data rows are Exent’s data. The lesson: compare a source’s “up” rate with how many stocks rose anyway, and its “beat the index” rate with roughly the share of stocks that beat the index in the same window. The large-stock figures exclude companies that shrank out of the list, so they flatter the base rate. The [2013 target-price study](https://ideas.repec.org/a/spr/reaccs/v18y2013i4d10.1007_s11142-012-9216-5.html) also shows why the horizon must be fixed: a target is far easier to touch once during a year than to hold at the end of it.

## Why does average excess return matter more than hit rate?

Because a source can be right most of the time and still lose money. What matters is the size of the wins against the size of the losses, measured above the index over the same dates. An illustrative comparison:

| Illustrative | Source A | Source B |
| --- | --- | --- |
| Hit rate (beat the index) | 70% | 40% |
| Average excess return when right | +5% | +30% |
| Average excess return when wrong | −20% | −8% |
| Average excess return per call | −2.5% | +7.2% |

Source A looks better on a hit-rate leaderboard; Source B made more money per call. Ask for both numbers, plus the typical drop along the way: a call that ends up right after falling 40% is one many investors would not have held. The [alpha](https://www.exent.ai/learn/alpha) entry covers excess return in more detail.

## How many calls before a track record means something?

Usually several hundred. With a few dozen calls, luck can produce almost any hit rate. The table shows the illustrative range a 60% hit rate could plausibly come from, at 95% confidence, if every call were independent:

| Calls in the record | Measured hit rate | Plausible true range (Illustrative) |
| --- | --- | --- |
| 20 | 60% | about 39% to 81% |
| 50 | 60% | about 46% to 74% |
| 100 | 60% | about 50% to 70% |
| 400 | 60% | about 55% to 65% |
| 1,000 | 60% | about 57% to 63% |

Real calls are not independent, because one market move lifts or sinks many picks at once, so the true uncertainty is wider. Real records behave the same way. In Exent’s analyst data, 52% of the 164 analysts with fewer than 100 rated calls landed below 30% or at 80% or more; among the 429 with 1,000 or more, only 5% did. A spectacular record over 20 calls is mostly noise. The analyst version of this test is in [how to tell if a stock analyst is any good](https://www.exent.ai/resources/judge-an-analyst).

## Is the record complete, and judged on a fixed horizon?

Only a complete, time-stamped record can be trusted. Three distortions are common:

- **Survivorship.** Sources with bad records close or rebrand, so the ones still advertising are the survivors. See survivorship bias.
- **Cherry-picking.** “Our best picks” pages show winners. Someone running several model portfolios, or posting many tips, can always find one that worked.
- **Moving horizons.** A call with no stated horizon can be declared right whenever the stock is up. US broker-dealers must define each rating, including its time horizon and benchmark, under [FINRA Rule 2241](https://www.finra.org/rules-guidance/rulebooks/finra-rules/2241); most tips have no such rule.

A fair record lists every call with its date, the price at the time, the horizon and the outcome, including calls that were later deleted or quietly dropped.

## How do conflicts of interest bias recommendations?

They tilt calls toward “buy” and toward stocks the source benefits from. The problem is documented at every level:

- **Banks.** In 2003, ten Wall Street firms agreed to a [$1.4 billion global settlement](https://www.sec.gov/litigation/litreleases/lr-18438) over research the SEC said was unduly influenced by investment banking. Rule 2241 now requires each rated report to show the share of the firm’s ratings that are buy, hold and sell, and how many of each are banking clients.
- **Promoters.** Anyone paid by a company to publicize its stock must disclose the payment and its amount under [Section 17(b) of the Securities Act](https://www.law.cornell.edu/uscode/text/15/77q).
- **“AI” claims.** In 2024 the SEC [fined two investment advisers $400,000](https://www.sec.gov/newsroom/press-releases/2024-36) in total for false and misleading statements about their use of artificial intelligence.

A conflict does not make a call wrong, but it changes how to read it. More in [are analyst ratings biased](https://www.exent.ai/resources/analyst-bias).

## Exent’s top 50 analysts

Across 988,127 ratings, the average Wall Street analyst’s price target was reached 53.0% of the time: close to a coin flip. The spread between analysts is wide, and 14.4% of the 971 analysts with a measurable record reached 70% or more (Exent data, Nov 16, 2026).

Exent Gems is built on the top 50 of Exent’s ranking. Over the last two years, some of them reached 90% or more of their price targets. Who they are, which stocks they are calling now and how often they have been right is for members.

[Unlock the top 50 analysts](https://www.exent.ai/radar)

## A checklist before you act on any recommendation

- Do I have the full, dated record, including the losers?
- How does the hit rate compare with the share of stocks that beat the index in the same period?
- Is the average excess return positive, or carried by one or two winners?
- Are there hundreds of calls behind it, across rising and falling markets?
- Was the horizon stated in advance?
- Is the source paid by the company, or holding the stock, and is that disclosed?
- Does anything survive fees, trading costs and taxes?
- Do other sources with strong records agree?
- How would the stock fit what I already own?

Checking a single tip from a forum, a friend or a newsletter? The step-by-step version is [how to check a stock tip before you buy](https://www.exent.ai/resources/how-to-vet-a-stock-tip); for Wall Street ratings, see [stock analyst ratings explained](https://www.exent.ai/resources/stock-analyst-ratings-guide). This is general education, not personal advice.

## How to check recommendations with Exent

1. Stocks Radar ranks thousands of Wall Street analysts by past accuracy, so you can see whose calls have held up before you read the next one.
2. Each analyst profile shows hit rate, number of ratings and average return after calls, next to their price targets.
3. Gems are stocks where several 4★-and-above analysts agree: a consensus of proven records instead of a single call.

Exent is our product: [Exent Gems](https://www.exent.ai/radar)

## Frequently asked questions

### How do I check if stock picks are good?

Get the full dated list of picks, then compare how often they beat the S&P 500 and by how much on average with what the index did over the same dates. Check the number of picks, whether losers were removed and whether the source was paid or holding the stocks.

### How do I judge an analyst’s track record?

Look at the price-target hit rate against the sector median, the number of ratings behind it, the recent record and how the rated stocks did afterwards. Hundreds of ratings over several years are needed before the record says much.

### What is a good hit rate for stock recommendations?

One clearly above the base rate for the same period. In Exent’s backtest, the average analyst’s price target was reached 53.0% of the time; for picks judged on beating the index, compare with the share of stocks that beat it anyway, which is often well under half.

### How many calls before a track record means something?

Usually several hundred. With 50 calls, a 60% hit rate could plausibly reflect a true skill anywhere from about 46% to 74%, so short records are mostly luck.

### Are AI stock ratings more reliable than analyst ratings?

Not by default. Most AI and quant ratings are judged on the vendor’s own backtests, which are hypothetical, and the same seven tests apply: benchmark, excess return, sample size, horizon, completeness, conflicts and costs.

### Can I trust a source that shows only its best picks?

No. A list of winners says nothing about the hit rate or the average return, because the losing calls are missing. Ask for every call with its date.

## Sources

1. [SPIVA U.S. Year-End 2025](https://www.spglobal.com/spdji/en/spiva/article/spiva-us-year-end-2025/) — S&P Dow Jones Indices
2. [Can Investors Profit from the Prophets? Security Analyst Recommendations and Stock Returns](https://ideas.repec.org/a/bla/jfinan/v56y2001i2p531-563.html) — Journal of Finance (Barber, Lehavy, McNichols, Trueman), 2001-04-01
3. [Do sell-side analysts exhibit differential target price forecasting ability?](https://ideas.repec.org/a/spr/reaccs/v18y2013i4d10.1007_s11142-012-9216-5.html) — Review of Accounting Studies (Bradshaw, Brown, Huang), 2013-12-01
4. [Finfluencers (working paper 23-30)](https://www.sfi.ch/en/publications/n-23-30-finfluencers) — Swiss Finance Institute (Kakhbod, Kazempour, Livdan, Schürhoff), 2023-04-25
5. [Rule 2241: Research Analysts and Research Reports](https://www.finra.org/rules-guidance/rulebooks/finra-rules/2241) — FINRA
6. [Federal Court Approves Global Research Analyst Settlement](https://www.sec.gov/litigation/litreleases/lr-18438) — U.S. SEC
7. [15 U.S. Code § 77q (Securities Act Section 17)](https://www.law.cornell.edu/uscode/text/15/77q) — Legal Information Institute, Cornell Law School
8. [SEC Charges Two Investment Advisers with Making False and Misleading Statements About Their Use of Artificial Intelligence](https://www.sec.gov/newsroom/press-releases/2024-36) — U.S. SEC, 2024-03-18

---

Educational content, not investment advice. Past performance does not guarantee future results.
