
投机实验室
投机实验室
·
我要验牌。
10Following
348followers
Feed
Livestream
Feed
An old strategy explained in just a few sentences, run on the S&P ETF for 33 years with a 77% win rate. Pretty impressive, right? 🥸😘
In the past two weeks, Bitcoin dropped from 77,000 to 62,000. It sold before the big drop and didn’t touch a single trade afterward. This was publicly shared by top U.S. trader Larry Connors, called the RSI 2 pullback buying method.
He’s the same guy who co-authored a book with market wizard Linda Raschke.
The idea is simple: in an uptrend, if the price falls too much for two consecutive days, buy; when it bounces back, sell.
First, look at the big picture: only trade if the closing price is above the 200-day moving average. Then check RSI 2, which only looks at the strength of price changes over the last two days. Buy at the close when RSI drops below 10, and sell at the close when the price closes back above the 5-day moving average.
He doesn’t set stop losses because his tests showed adding stop losses reduced profits. So position sizing and diversification are important.
But it doesn’t earn as much as just holding. Over 33 years, the S&P averaged 5.2% per year with this method, versus 10.7% just holding.
The advantage is it’s only active for a bit over a month each year, with a max drawdown of 15%. Holding the market saw max drawdowns of 55%. No stop loss has its cost; you have to endure big drops. For example, Bitcoin had a single trade loss of 22% in May 2021. Never use leverage with this.
I summarized all the backtested numbers and the full trade process from buy to sell in the chart.

Urgent update, those who have crypto really need to be careful, this is very common, I've encountered it myself 😡🥸
For example, if you copy a crypto address from big sites like Zoom, BBC, CNN, the address might have been replaced by a hacker's from the afternoon of the 16th to the morning of the 17th. A total of 178 websites were affected, including the official Ledger site in the crypto space!
These websites themselves were not hacked; they all use a web testing tool called Optimizely. Each page pulls a piece of code from its server, and the websites don't review it themselves.
Hackers exploited a vulnerability in Optimizely's registration process and injected their own code into the projects of 178 clients. The code is sent from Optimizely's legitimate servers, so browsers don't detect any issues.
This code searches the page for crypto addresses and replaces them with real addresses that can receive coins, so wallets can't detect the change.
Once you copy, your clipboard contains the hacker's address, and if you transfer coins, they're gone. While that page is open, copying addresses elsewhere will also be replaced. By the evening, they added code to steal credit card numbers, and if you open it on an older iPhone system, it might even try to attack.
In the future, when transferring coins, after pasting into your wallet, verify the displayed address on the screen against the one the other party gave you from start to finish. Checking only the first and last few characters is not enough. If you use a hardware wallet, check the address on the hardware wallet's screen.
If you transferred coins between the 16th and 17th, double-check the recipient address afterwards.

This is the "Financial Stability Report" just released by the Federal Reserve, 69 pages, only posted as a PDF on the official website, no press release issued, and I couldn't find any reports in either Chinese or English. I have finished reading it.
The report states that the high valuation of the US stock market is largely due to market enthusiasm for superintelligence. There is a dedicated section in the report discussing AI giants borrowing money 🤔🤑
The debt issued by big companies for computing power was $130 billion in 2023 alone, and by October this year it has already exceeded $680 billion, a fivefold increase. Now, for every $3 borrowed by large companies, $1 is borrowed for computing power.
There is also a large amount not counted as liabilities.
Amazon, Google, Meta, Microsoft, Nvidia, Oracle, and Broadcom have signed leases not yet started totaling over $1 trillion, and various guarantees and contracts that must be paid regardless of use exceed $2 trillion.
The report also describes a situation where GPU manufacturers sell cards to AI companies and guarantee the future value of these cards. The big companies then invest money into these companies, and the money cycles back again. The Federal Reserve calls this circular financing.
People in the market are also worried about this.
The New York Fed surveyed 23 brokers, bankers, and fund managers, and 87% listed AI as the top risk, compared to only 30% a year ago.

This is the 10-month moving average timing strategy that fund manager Meb Faber wrote about in a 2006 paper. I ran it with real market data, checking only once a month. Over 33 years of the S&P, the maximum drawdown was only 25%, which is pretty impressive 🥸🧐
The rule is simple: on the last trading day of each month, check the closing price. If it's above the 10-month moving average, hold; if it falls below, sell everything and put the money into short-term government bonds to earn interest.
The 10-month moving average is the average of the closing prices at the end of each of the last 10 months. Mid-month price changes are ignored, and no stop-loss is set.
Here's how it works for Bitcoin: at the close of November 2025, the price was 90360. If it falls below the moving average, sell everything. By then, it had already dropped 27% from its peak, so it never sells at the highest point—it always drops first.
Afterward, Bitcoin fell to a low of 58625, which this strategy avoided. It returned above the moving average by the end of August 2026, buying back at 78581, which was 13% cheaper than the selling price.
The advantage is less loss; it doesn't necessarily earn more than just holding.
In the 20 years since the paper was published, this strategy earned on average 2.4 percentage points less per year than buy-and-hold on the S&P, checking only once a month. It can't avoid Bitcoin dropping by half in a month, and during sideways markets, it might sell just before a rise, only to buy back at a higher price.
The current Bitcoin position was bought at the end of August and has already lost 18%. The rule says to wait until the close on November 30; if it doesn't rise back above 71842, sell. 😂

This chart is very interesting. By time, Bitcoin's current bear market should have bottomed out.
The previous three cycles each bottomed out after about a year. Today marks the 408th day of this cycle, but in terms of depth, the previous three cycles fell at least 77%, while this one has only dropped 53% at its deepest. If it really falls 77%, that would be 29,000 🥹😢
It's like carving a mark on a boat to find a sword—kind of nonsense. People say all sorts of things nowadays.

Even Anthropic's official skill packs were flagged as cheats by Nvidia's scanner, isn't that ridiculous? 😂
I scanned 7 of the hottest skill packs with it; out of 134 skills, 20 were flagged as cheats. I opened the 5 most dangerous ones to check, and they were all normal operations. I didn't check the remaining 15 one by one...
Now AI like Claude Code and Codex can also install skill packs.
A skill pack is basically a manual plus some scripts; once installed, the AI follows it, having all the permissions on your computer. Nvidia has studied over 30,000 skills, and a quarter of them have vulnerabilities, so they open-sourced this scanner called SkillSpector, which already has 26,000 stars on GitHub.
It has two modes: one free that scans code based on rules, and one that connects to large models requiring an API key.
I used the free one. Among Anthropic's official 19 skills, 5 were flagged as cheats. The Claude stock trading skill pack I tested yesterday was also flagged.
Looking closer, the stock trading pack was flagged as stealing keys, but the code actually uses your own broker key to check your account on the broker's official website.
Anthropic's PPT skill deletes the previously generated PPT and repackages it, and it was also marked as a dangerous operation.
So, you can be cautious about cheat and high-risk flags, but don't trust them blindly.

Anthropic hasn't set a price yet, but its perpetual contract on OKX has already risen to $98. The underwriters are signaling a price above $80; if it really sets at $85, that means before the market even opens, the perpetual contract has already increased by 15% 😂
Most of this jump is thanks to Nvidia.
Nvidia's earnings report last night was a threshold before Anthropic's pricing, with earnings per share at $2.68 and Q4 guidance at $130 billion, both higher than Wall Street estimates, leading to a 4.8% rise after hours.
This afternoon, the perpetual contract rose from just over $96 to $98. Bloomberg says the underwriters are guiding the price toward the top of the range or even higher, and the Wall Street Journal says at least $80.
So on day one, what I'm more concerned about is how much room is left after the perpetual contract has already risen. If priced at $85, a 20% increase would be $102, which is only 4% above the perpetual contract. Regarding market cap, based on the number of shares issued and private placements in the prospectus, closing around $95.9 would be about 1.25 trillion.
The perpetual contract is only 2% higher than that, meaning I expect the closing price on day one to be mostly below $102, with gains within 20%, and market cap around 1.25 trillion.
If the first trade opens above $105 or the close stands above $102, then I have underestimated it. The pricing will be announced between 7 and 10 AM tomorrow, and I will follow up immediately on my OKX Orbit.


Anthropic will price tomorrow morning, and OKEx perpetual has already bet on more than 10% of the first-day increase, it will probably go like this tomorrow.
Anthropic will only set the price tomorrow morning, but its perpetual contract on OKX has already been traded up to $96.83. This price is as of 2:26 PM Beijing time on the 18th. I estimate the pricing will be around $85, so by this calculation, those buying the perpetual contract have already bet on a 13.9% increase on the first day. OKX's perpetual contract is priced based on the number of shares issued by Anthropic. On November 12, OKX performed a rebase. The number of shares used for pricing was changed from the originally estimated 10 billion shares to 13.116 billion shares according to the prospectus. The announcement clarified that this is not a fully diluted share count. The contract price multiplied by this share count corresponds to its valuation, which is currently $1.27 trillion. Looking at the valuation, this perpetual contract has cooled down quite a bit over the past month. In mid-October, it was still around $2 trillion. After the prospectus was made public on October 26, it declined steadily, reaching $1.34 trillion on November 6, and now stands at $1.27 trillion. The tone for pricing moving upward is very clear. Bloomberg reported on Tuesday that underwriters have already guided the price to the top of the range or higher, with several orders exceeding $10 billion. The Wall Street Journal says the minimum is $80. But the prospectus still shows $70 to $78, and on the same day, it filed registration documents for listing on Nasdaq, with pricing set for tonight. My original biggest concern was Nvidia's earnings report, and that hurdle has been cleared. Nvidia's adjusted earnings per share were $2.68 for Q4
South Korean legendary trader Kim Hyung-jun earned between 15 billion and 20 billion KRW through trading, roughly 15 million USD. His method relies on nothing else but an extremely high win rate 😂🥸
Out of over 200 trading days a year, he only loses money on about twenty or thirty days. Over fifteen years of full-time trading, calculated monthly, he only lost money in two months.
His strategy is called the New High Method, and I have made the entire strategy into this one chart.
The logic is to wait for a stock to hit a new high, then pull back a bit before buying. Once it hits a new high, there are no longer people stuck waiting to break even and sell. But most new highs are driven by a big bullish candlestick, so chasing on the same day means buying at the highest point.
First, stock selection: he trades short-term, looking at the top 50 stocks by daily trading volume, excluding ETFs, picking leading stocks linked to new themes. When they hit a new high, he adds them to his watchlist, and the stock price must also be above the 120-day moving average.
Then he waits for the buying point: he looks for a rising bearish candlestick, which is a bearish candle formed during an upward price movement. The trading volume during the surge must exceed 50 billion KRW, the bearish candle’s close must be above the 5-day moving average, and institutions must still be buying that day. When all three conditions are met, he buys before the close.
If the price fails to rise after the market opens the next day, he immediately cuts losses. The amount bought each time is predetermined as a percentage of total capital, and he concentrates on only two or three stocks at a time.
He holds as the price rises and sells once it falls below the 5-day moving average. If there is a sudden surge in volume more than five times the usual during a high-level rise, he treats it as a real top.

Dan Bin said this year is like 1994, not like 2000😂🥲
I pulled up the Nasdaq to check, and this round this year is only 2% away from the deepest drop point in 1994.
In 1994, the Nasdaq peaked in March, then fell all the way until the end of June, dropping as much as 13.7%, closing the year down 3.2%, and rising 40% the following year.
This year, the peak was on October 6, and as of last night, it has been exactly 6 weeks with an 11.9% drop.
In 2000, the peak was on March 10, and after the same 6 weeks, the Nasdaq had already dropped 28%, and by the end of the year it was halved. Judging by the speed of the drop, this year is on par with 1994, far from 2000.
But 1994 closed the year with a loss. Last night, the Nasdaq closed at 24,328, 4.7% higher than the end of last year. To close the year down, it still needs to drop another 4.5% before year-end.
If the 23,800 line breaks, this drop this year will be deeper than the deepest point in 1994.



