On July 30, IBM chief executive Arvind Krishna joined Jim Cramer on Mad Money to talk about quantum computing. When Cramer asked how soon quantum machines could threaten the cryptography protecting digital assets, Krishna told him to start getting cautious within three or four years.
Cramer heard a countdown moving much faster. The next day, while discussing the interview, he said, “I’m going to sell mine,” referring to Bitcoin. He had gone from asking about a future technical risk to planning an exit.
Crypto traders knew the joke before the clip finished spreading, and their reflex was to buy.
Bitcoin closed July 31 at $62,813. Less than three weeks later, the market broke upward. Bitcoin closed at $69,266 on August 19 and $73,033 on August 20. In the following week, it traded as high as over $80,697. Anyone who bought the July 31 close was up 28.5 percent at that point.
Screenshots soon paired Cramer’s sell call with Bitcoin’s rising chart.
Three examples can keep a meme alive. A trading rule has to account for the calls nobody reposts.
I wanted to know whether “Inverse Cramer” could survive contact with a ledger, so I dug deeper. Below is what I found.
The trade that sounded too easy
The internet’s version of Inverse Cramer requires almost no thought. When Cramer says buy, you sell. When he says sell, you buy. His loudest misses become proof that the rule works.
Bear Stearns became the foundational clip. Bitcoin supplied several sequels. Nvidia, Meta and a rotating cast of meme stocks supplied the rest. Each episode followed the same editing logic: isolate a televised opinion, jump forward to a painful chart and let the audience fill in the conclusion.
Once actual money enters the picture, the rule needs definitions. A backtest must decide which comments count, whether entry occurs at the close or the next open, how long positions stay open and how repeated mentions are handled. A short strategy also has to account for market hedges, borrow fees and dividends.
Wall Street eventually packaged the meme into a security. The Inverse Cramer Tracker ETF, ticker SJIM, began trading on March 1, 2023. Its SEC prospectus said the fund would monitor Cramer’s television programs and social-media comments, then take the opposite side of his stock, sector or market views. It expected to hold 20 to 50 positions, trade frequently and charge 1.20 percent in annual operating expenses after the stated waiver.
Investors could buy the joke with one click, but the fund lost money. An SEC shareholder report recorded a 5.04 percent loss from inception through August 31, 2023. Over the same period, its Nasdaq 100 total-return benchmark gained 30.39 percent. Adjusted market-price data through SJIM’s final exchange-trading date show a loss of about 15.7 percent from March 1, 2023 through February 13, 2024. SPY gained about 27.1 percent over those same dates, while QQQ gained about 48.2 percent.
The Long Cramer Tracker ETF, LJIM, did better than its inverse sibling but still failed to keep pace. The SEC report recorded a 6.20 percent NAV return through August 31, 2023, against 30.39 percent for the Nasdaq 100 total-return benchmark.
Both funds closed. LJIM stopped trading on September 11, 2023 and liquidated on September 21. The SJIM board approved liquidation the following January, with February 13 set as its last exchange-trading day. The SEC liquidation notice says the board concluded that closure served shareholders’ best interests. It does not give us a clean causal verdict on why assets failed to gather.
Investors who bought the blanket inverse lost money while SPY and QQQ rose.
Television moves the opening price
Research on the show’s price impact starts with the overnight gap, before an ordinary viewer gets a fair chance to trade.
Joseph Engelberg, Caroline Sasseville and Jared Williams studied 826 first-time buy recommendations broadcast between July 2005 and February 2009. Their paper, “Market Madness? The Case of Mad Money”, found an average abnormal overnight return of 2.4 percent after a recommendation. That translated into an average $77.1 million increase in market value before the next regular session.
The median overnight move was smaller, 1.18 percent, because a handful of recommendations produced huge jumps. The pattern grew stronger among small and illiquid companies. These were the stocks where a nationally televised mention could send a concentrated wave of retail demand into a thin order book.
The bump then faded. Portfolios formed after the recommendation, once the first tradable opening price had arrived, produced annualized alpha of negative 9.98 percent at the 50-trading-day horizon, negative 6.15 percent at 150 days and negative 3.2 percent at 250 days. The stocks with the largest opening jumps reversed the hardest.
Short sellers noticed the same distortion. A separate study of 1,234 Mad Money buy recommendations found unusually heavy short selling after Cramer’s calls, followed by price reversal. Short sellers were leaning against the attention spike while the new audience was still arriving.
The pattern barely appeared after sell recommendations. Engelberg and his coauthors measured an average abnormal overnight return of negative 0.29 percent after first-time sells, far smaller than the reaction to buys, with no detectable post-recommendation trend. Retail investors can buy a newly discovered stock with a few taps. Acting on a sell recommendation requires them to own it already or to borrow shares. The machinery is asymmetric.
Paul Bolster and Emery Trahan found the same split in their 2009 study, “Investing in Mad Money”. Cramer’s calls moved prices, and the buy effect reversed, but his average recommendations were neither extraordinarily good nor unusually bad. Factor exposure also shifted across periods.
His longer-running Action Alerts PLUS portfolio also lagged the market without turning into an obvious short. Jonathan Hartley and Matthew Olson studied its history from August 2001 through March 2016. The portfolio gained 64.45 percent, compared with 126.06 percent for the S&P 500 total-return index. Annualized returns were 3.38 percent versus 5.59 percent, with Sharpe ratios of 0.11 and 0.24. The CAPM estimate showed negative annual alpha of 2.38 percent at the 10 percent significance level, although much of the gap came from lower market exposure and cash held for charitable distributions. Adding more factors removed the statistical significance in several specifications.
Lagging a benchmark does not automatically make a profitable short. A portfolio can rise while trailing SPY. Anyone who shorts it may lose money even though the relative-performance chart looks terrible.
Sixteen thousand calls change the picture
A 2026 working paper gives us the largest recent test I found. Andres Kull assembled 16,701 long-side recommendations extracted from Mad Money broadcasts between January 2018 and December 2024. The study enters at the next trading day’s open, measures returns against SPY over fixed horizons, groups repeated mentions into sequences and uses ticker-clustered errors so 179 Nvidia mentions do not masquerade as 179 independent discoveries.
The blanket inverse failed again. Across 8,169 one-year observations, a market-neutral trade that shorted Cramer’s picks and bought SPY produced average per-position alpha of negative 0.34 percent. The p-value was 0.57. Statistically, the result was indistinguishable from zero and pointed in the wrong direction.
Results changed sharply with company size. Cramer’s casual mega-cap buys beat SPY by 6.8 percent over the following year in the study sample. His casual large-cap calls were mildly positive. The weakness appeared as market capitalization fell. Mid-caps trailed SPY by 1.9 percent. Small-cap casual buys, defined in the paper as companies below $2 billion, trailed SPY by 24.5 percent. Seventy-nine percent underperformed the index.
The small-cap result depended on how the position was constructed.
A simple short of those small-cap calls earned an average 11.9 percent in absolute terms because the stocks themselves lost 11.9 percent. SPY gained 12.7 percent during the same windows. Compared with buying the index, the naked short still lost 0.8 percentage points, with a p-value of 0.85.
The market-neutral pair worked differently. Short the recommended small-cap stock and buy an equal amount of SPY. That combination earned 24.5 percent per position in the historical sample, the inverse of the stock’s SPY-relative loss. The ticker-clustered p-value came in below 0.0001.
Volatility then split the result again. When the VIX was below 30, 350 qualifying small-cap casual buys produced average pair-trade profit of 28.7 percent over one year. In the 31 High-VIX cases, the pair trade lost 22.1 percent. A naked short during those stressed periods lagged SPY by 82.2 percent as beaten-down small companies snapped back.
The historical setup was narrow: a first-time casual buy in a small company, entered at the next open, hedged with an equal SPY position and avoided when the VIX exceeded 30. It appeared about 50 times a year in the 2018 to 2024 sample. The result remained strong across a grid of market-cap and VIX cutoffs, even after the author modeled stock-borrow costs as high as 3 percent.
The same data argue against inverting Cramer’s mega-cap casual buys, which outperformed SPY. Recommendations tied to companies he had previously disclosed as personal charitable-trust holdings showed different behavior from one-off calls. Treating every mention as equal throws away the categories that explain the returns.
Then the broad 2025 test fought back
The working paper’s 16,701-call result made a blanket inverse look dead. Our next test used a later forward period and a simpler question: among mature, direct stock-buy recommendations, did buying SPY and shorting the recommended stock work over the following year?
This was a retrospective forward-period test, not a strategy we traded live. The 2025 calls and their one-year holding periods had already happened. We did, however, keep the final quote decision away from the returns. Two independent reviewers saw recommendation wording without prices, market values, VIX readings, entry dates or exit dates. A call entered the primary set only when both reviewers marked it as a direct recommendation.
The funnel began with 393 eligible Mad Money episodes. A deterministic source pass found 361 high-confidence events, including 266 explicit buy candidates and 245 identity-resolved stock buys. After same-episode duplicates and incomplete outcomes, 78 mature candidates reached the quote review. The reviewers agreed on 72. Six disagreements were discarded. Forty-two direct recommendations across 29 tickers survived.
The result favored the meme. The equal-dollar pair, long SPY and short Cramer’s recommended stock, gained 10.35 percentage points on average before costs. The median was 11.29 points, the inverse won 64.3 percent of the time and the two-sided p-value was 0.0173. A ticker-clustered bootstrap put the 95 percent interval between a 0.92-point gain and a 19.18-point gain.
Repeated recommendations did not create the result. Grouping calls within 60 days and keeping only sequence starts left 33 events, with a 10.62-point mean and a p-value of 0.0407. Nor did one ticker carry the sample. Dropping each ticker in turn left the mean between 8.22 and 12.16 points.
Simple cost assumptions weakened the edge without erasing it immediately. With 50 basis points of borrow and 20 basis points of execution, the mean fell to 9.65 points. At 100 plus 20, it was 9.15. At 300 plus 20, it was still positive at 7.15 points, although the p-value rose to 0.0943.
I reran the classifications and concentration checks before trusting this result. The ETF had failed, and the largest historical dataset found no blanket edge. Yet a later 42-call sample, reviewed without outcomes, landed on a positive estimate with an interval above zero.
It still falls short of a trading mandate. Forty-two events are enough to be interesting, not enough to settle a meme built from thousands of calls.
The small-cap filter still failed
The 28.7 percent figure comes from a working paper whose strongest thresholds were selected after the author inspected the data. The upstream signal-extraction pipeline is absent from the repository, and its market-cap buckets use a current snapshot rather than each recommendation date. A result can look excellent inside its discovery sample and vanish when the calendar moves forward.
We reconstructed that conditional trade separately on mature 2025 Lightning Round calls. The rule stayed narrow: explicit bullish recommendations, no Charitable Trust holdings, company value below $2 billion at the next open, VIX below 30, and a first qualifying call after at least 60 days. Each event shorted the recommended stock, bought an equal dollar amount of SPY and closed both legs after one year.
This source review started with 774 CNBC recap calls, a different universe from the broad test. We sent 552 common-stock quotes through two independent, return-blind reviews. The reviewers agreed on 539 and disagreed on 13, which the strict rule excluded. That left 215 consensus bullish calls, 200 after removing Trust-linked names, 18 small-cap candidates and 17 final events after the VIX screen. The final ledger was frozen before outcomes were joined.
Those 17 recommended stocks gained 42.5 percent on average. SPY gained 20.3 percent over the matched windows. The inverse pair lost 22.2 percentage points per event, almost the mirror image of the paper’s positive 28.7-point estimate. Its median result was positive 0.6 points and nine of 17 events won, but a few large stock gains, including Powell Industries, Arcus Biosciences and Lincoln Educational Services, crushed the mean.
The p-value was 0.293. A ticker-clustered bootstrap put the 95 percent interval between a 66.2-point loss and an 18.7-point gain. After a simple 1.2 percent cost deduction, the mean fell to negative 23.4 points. Removing the acquired Office Depot observation, keeping only the first call per ticker, or weighting tickers equally left the result negative.
The broad test and the conditional reconstruction do not cancel each other out. They test different source universes and different rules. Together they show how quickly “Inverse Cramer” changes when recommendation wording, ownership, company size and volatility enter the definition. A profitable result from one slice cannot authorize a strategy built from another.
What we learned
The blanket version of Inverse Cramer has weak support. The live ETF lost money, and the 16,701-call study found no broad advantage from reversing every recommendation.
Cramer’s show can still move prices. The older studies found a real attention effect, especially in small companies where a wave of retail buying can push the opening price higher. That effect helps explain why individual Inverse Cramer examples look so persuasive.
Our 2025 results show why the meme remains hard to dismiss. The broad sample produced a 10.35-point inverse gain. Yet the supposedly strongest historical rule failed when applied to a later group of small-cap recommendations.
The difference probably comes from what gets counted. Recommendation wording, company size, ownership, market conditions and the source used to collect calls can all change the sample. A profitable result in one group does not automatically carry into another.
Forty-two broad calls and 17 small-cap calls are also limited samples. Public recap pages miss some spoken comments, and historical shorting costs are difficult to reconstruct stock by stock. The numbers are strong enough to challenge the easy verdict, but too unstable to support a live allocation.
I started this project expecting one of two clean answers: the meme works, or the meme is nonsense. We found neither.
Inverse Cramer may contain a market signal. We still do not have a dependable trading rule. Capital stays out.
Disclosure: This article is for informational and educational purposes only. It is not investment advice. Short selling can produce unlimited losses, and market-neutral trades carry borrow, execution, tracking and liquidity risks.