Even with a 95% win rate, I still lose money: I've already fallen into the trap of "financial management" schemes that predict the market's closing bell.

golem

Odaily资深作者

@web3_golem

2025-11-11 12:17

This article is about 2728 words, reading the full article takes about 4 minutes

From earning 1% every day to losing so much that I questioned my existence.

Original article | Odaily Planet Daily ( @OdailyChina )

Author｜Golem ( @web3_golem )

This year has seen a rollercoaster ride in the crypto market, with narratives of memes, DAT, AI, and stablecoin investments rotating repeatedly, but each only enjoyed a brief moment of spotlight before being ousted. In contrast, only the prediction market has consistently held center stage.

The high degree of product-market fit of prediction markets is gaining increasing recognition. In November, Google announced that it would directly integrate Polymarket and Kalshi prediction market data into Google Finance. Trump's social platform, Truth Social, also announced the upcoming launch of a new service called "Truth Predict," which allows users to use cryptocurrency as betting capital to predict the outcome of specific events, covering areas such as sports games, political elections, and even economic changes.

Meanwhile, Polymarket, the leading prediction market that has completed over $2 billion in funding and is valued at $9 billion, has confirmed that it will airdrop tokens in the future. Therefore, with the narrative gradually heating up and the airdrop confirmed, Polymarket has undoubtedly become the next key project for profiting from token drops. Many users who do not usually participate in prediction markets are also looking forward to obtaining future "big profits" through interactions with the least possible loss.

However, market predictions also follow the Pareto principle: most users will lose money .

Against this backdrop, many KOLs have begun recommending "end-of-day strategies" to retail investors. These strategies involve betting on events where the outcome is nearly certain, with a probability of 95% or even 99% or higher, aiming to capture the final few percentage points of guaranteed profit. Some have even gone so far as to treat "end-of-day strategies" as a new type of financial tool, claiming that " if you achieve a 1% daily return, the annual compound interest can reach 365% ."

The ideal is beautiful, but reality can be harsh. The so-called "end-of-day strategy" is not as safe as KOLs claim. You may have a high win rate, but you will still suffer losses in the end.

The "end-of-day strategy" had a 95% win rate, but still resulted in losses in the end.

For a while, I was convinced of the magic of "end-of-day strategies" on Polymarkets. This stemmed from my fundamental belief in prediction markets like Polymarkets: people make responsible predictions due to vested interests, and the probability of betting tends to align with the probability of actual events occurring through rational analysis. Under these circumstances, the probability of losing money by sweeping the end of the day is extremely low.

At the same time, I not only manually executed the "end-of-day strategy," but also, to completely free up my hands, I delegated the execution of the strategy to an AI agent, which searched for events with a probability greater than 95% and placed bets (related reading: Can using an AI agent to execute the end-of-day strategy get you a Polymarket airdrop? ). At the time, I thought I was incredibly clever, as it automated the interaction to obtain potential airdrop eligibility and was also a stable financial management method. In hindsight, this was a very foolish decision.

Out of caution, I only wanted to use $100 for the strategy, but after testing for 9 days I decided to stop running the AI agent .

This AI agent made a total of 146 bets during its operation, with 16 bets per day from tomorrow. Anyone who has actually played prediction markets knows that this frequency is far higher than that of ordinary people. Furthermore, according to Polymarket Analytics data, the AI agent had a win rate of 95.1%, but only earned $3, incurring a loss of $16. Ultimately, by the time I stopped running it, the total loss was $12.

You might think that a $12 loss or an overall -12% return doesn't seem serious, and wonder how I can't even afford to lose that much money? That's because I realized the problem with the "end-of-day strategy" before it became serious (-80% return).

In over 100 bets, my AI agent only suffered losses on two events : "New York City's highest temperature on October 30th" and "Who Trump will endorse for New York City mayor." Dramatically, because before Trump endorsed Andrew Cuomo, the probability of "not supporting Andrew Cuomo" and "no one will endorse him" on Polymarket was both above 95%, so the AI boldly bet on both outcomes simultaneously, ultimately losing on both.

Before using AI to execute the closing strategy, I was aware of the "black swan" risk of the closing strategy. In order to deal with this risk, the AI agent also made a split betting design, that is, only 5% of the position is bet each time (only $5 is bet each time with a principal of $100).

However, I overlooked the frequency of AI-managed transactions. A dozen or so bets per day would amplify the probability of a "black swan" event. The more bets made, the more likely it is to make mistakes. A human might not encounter a "black swan" event at the end of the trading day even once a month, but the AI encountered it three times in nine days.

I realized that losses accumulated faster than revenue, which was the main reason I stopped using this AI proxy strategy. Even if the AI only bets 5% of its position each time, it would usually make a 1% profit. However, once it incurred a loss, the AI would need to bet on the closing bell and win 103 times consecutively to recover the original 5% loss. In reality, due to the frequency issue, it's impossible for the AI to avoid encountering a "black swan" event at least once in those 103 attempts. Therefore, my AI proxy might never be able to recover the losses and could only continue to accumulate them.

This is a simple math problem, and with the help of ChatGPT, I figured out the risk hidden behind the high odds of the end-of-game strategy in 5 minutes: it's a foolish strategy that only allows winning and betting big on small wins .

On UnifAI, 379 wallets are using AI agents to employ end-of-day strategies, with a total fund size of $59,000. However, the ROI of this strategy so far is -44.93%. I hope users who read this article will think carefully about whether your expectations for airdrops outweigh your accumulated losses.

Trying to gamble with a large investment on a small one will ultimately result in being wiped out in one fell swoop.

What about manual operation or reducing the frequency? Would that prevent us from making the same mistakes as AI?

This idea is naive. No matter how smart, careful, or adept I am at controlling my luck, I will eventually be swept away in one fell swoop when using a strategy of betting big against small.

On November 9th, a Polymarket trader in the top 0.7% of profitability was wiped out overnight, going from a $3,500 profit to a $15,000 loss (related reading: Polymarket Top 0.7% Winner's Account: How to Lose $19,000 in 1 Minute ). Perhaps no one understood end-of-day trading strategies better than him, as he had used this strategy to earn $3,500 over the past month, with a total trading volume of $200,000.

This trader was betting thousands or even tens of thousands of dollars each time, hoping to make a "guaranteed return" of 1% or less. The result? He was wiped out on the night of November 9th by a game that turned the tide. If he continues to use end-of-game strategies to try and recover, his performance won't be much better than my AI agent.

Essentially, prediction markets like Polymarket are gambling platforms. The correct approach is always to aim for high returns with low stakes, focusing on alpha rather than beta strategies. True market savvy players don't advocate end-of-day strategies. As Lone Wolf Capital ( @AnselFang ), who earned $2.93 million on Polymarket in 25 days, stated, ensuring a high degree of certainty before pursuing a small end-of-day gain, ultimately resulting in a loss of principal, is always a losing proposition.