The popular joke on X is that Jim Cramer is countersignal.
But what does the data actually show? Is that actually a profitable strategy?
And zooming out .... every investor, whether we admit or not, get many trade ideas from X.
I know several portfolio managers who’ve done extremely well by knowing whose ideas to follow.
There's a few challenges to copytrading well:
Was this call just lucky? Or has this individual been consistently correct?
What kinds investments is this account particularly good at? Eg., top hit rate on macro? Commodities? Tech?
With AI, we can easily turn this into a data science exercise.
And this week I'm sharing with you an AI skill to backtest any X account to figure out what the track record is .... which ultimately lets us know if it's an account worth listening to.
For our demo below, we'll be using Jim Cramer to answer our question "Should you really trade against Jim Cramer?"
Download my skill to follow along:
How I built the backtest
How I built the backtest
The process is pretty simple:
Pick an X account and a time period.
AI reads the available posts and identifies bullish or bearish calls. Celebrating last week’s rally doesn’t count as a prediction. Conditional calls stay flagged for review.
It groups the reasoning: valuation, fundamentals, momentum, etc.
Python calculates what happened next. Every call links back to the original post so you can check the AI’s interpretation.
For this test, we measured price changes from the next trading day’s open to roughly one month later.
(Tip: adjust the methodology if you trade different time horizons)
Then we asked two questions:
Did the stock move in the predicted direction?
Did it outperform or underperform the S&P 500, using SPY as the benchmark?
Example: someone says buy. The stock rises 2%, but SPY rises 5%.
Right direction. Still behind the market.
For bearish calls, we separately checked whether the stock fell and whether it underperformed SPY.
Let's get into the takeaways....
1. His one-month predictions were basically 50-50
His views got the direction right 50% of the time. They were correct against SPY 48% of the time.
The 50-50 result needs some market context. Most of his views were bullish, and bullish predictions can benefit from a rising market.
I applied the same bullish and bearish directions to SPY over the same dates. That got the direction right about 56% of the time. His individual stock views did worse on that measure.
The average result relative to SPY was essentially flat.
Following his calls didn’t show a dependable advantage in this test. Reversing them didn’t produce a positive average price result either. A less satisfying answer for the meme, perhaps, but that’s what the numbers showed.
2. His valuation calls hit 56%. His momentum calls hit 45%.
Calls based on whether something looked cheap or expensive performed better than calls based on momentum.
The comparison covered about 100 valuation calls and 140 momentum calls. “Valuation” here describes his reasoning in the post, rather than whether the company belongs in a value-stock portfolio.
Figma was a good example of what those valuation comments looked like. In July 2025, Cramer warned against placing a market order, pointing to its high sales multiple. From our measured entry, the stock fell about 49% over the following month. That warning held up.
Momentum was weaker. I also checked whether a longer wait improved those results. It didn’t do so consistently, which makes it hard to explain the misses as simply being early.
This was more useful to me than the overall 50-50 score. Perhaps it's worth listening when Cramer is talking about valuation!
3. 2022 was his weakest year in this sample
Only 41% of his 2022 views got the direction right over the following month.
Splitting the record by year made 2022 stand out immediately. The other years were much closer to 50-50, with hit rates between roughly 51% and 54%.
The 2022 calls also averaged about 2 percentage points worse than SPY over their measured month. That is an average outcome for individual calls, rather than an annual portfolio return.
It was also a difficult backdrop for his mostly bullish views. Applying his call directions to SPY produced a low hit rate that year too, about 37%. His chosen stocks still lagged SPY on average.
I suspect a year like this helped reinforce his reputation as a counter-signal.
4. Counting repeated calls changed the result
His hit rate against SPY fell from 48% to 46% when we reduced the weight of repeated views.
Imagine an account is bullish on the same stock on Monday, Tuesday and Wednesday. You have three posts, but they may all express one continuing investment idea. Counting all three gives that idea more weight than a stock mentioned once.
We checked this two ways. First, we kept the earliest call and suppressed further calls on the same stock in the same direction for about a month. Second, we gave each security equal weight, so frequently mentioned stocks couldn’t dominate the result.
Both produced a hit rate against SPY of about 46%. The average result also moved from essentially flat to slightly below the market.
This is easy to miss when an account has a huge feed. A record of a thousand posts can contain far fewer distinct investment ideas. In Cramer’s case, reducing that extra weight made the result look a little worse.
5. The average win and loss were both about 10%
Correct calls averaged about +10%. Incorrect calls averaged about −10%.
The individual calls were much less boring than the overall score. Hit rate tells you how often someone was right, but it leaves out how much the asset moved.
A bullish CoreWeave post in May 2025 preceded a gain of more than 160% over the measured month. An Intuit buy call in January 2026 preceded a decline of about 37%.
Across the full sample, the average gain on correct calls and loss on incorrect calls were similar in size. The average signed price result was a small positive, less than 1%, before costs. Relative to SPY, it was essentially flat.
These are outcomes for individual views. We didn’t build a portfolio with position sizes, overlapping holdings, or trading costs, so they don’t tell us what an investor would have earned by following his account.
Try the skill on an account you follow
The reason I built the skill was to make it easier to check anyone whose investment views I might follow.
The next time someone’s thread makes you want to buy a stock, you can look at their record: their hit rate, performance against the market, and which kinds of calls held up. The report also checks whether repeated posts change the results.
Check it out here:
-Dave
Personal
Join me live this coming Thursday, October 15, at 12pm ET for a (free) webinar with the Alphasense team!
The event will be recorded too so if you'd like a replay sign-up and you'll get the video to your inbox afterwards.
I'll be covering:
Data from a survey of 1,000+ AI decisionmakers on how their companies / funds are using AI
How you can get the most out of AI tools
My stories of what I've seen the top AI users leverage these tools for
