저자: Lumiwealth채널진행자 · 외국인
요약 보기
이 영상은 AI 에이전트 5개가 서로 토론하며 운용하는 자동 투자 전략을 소개한다. 전략 개요: 성장, 인플레이션, 부채·유동성을 분석하는 세 개의 분석 에이전트가 각자 근거를 제시하고, 반대 의견을 내는 에이전트가 어떤 주장을 따를지 판단하며, 실행 에이전트가 주문과 리스크 관리를 맡는다. 레이 달리오식 매크로 투자 스타일을 모방한 구조라고 설명한다. 성과: 2025년 7월 29일 약 10만 달러에서 시작해 현재 약 11만1천 달러로 S&P500(약 105 수준) 대비 약 3배에 가까운 수익률을 기록했다고 주장한다. 대폭락장 속에서도 버텼다고 강조한다. Python 대비 장점: 기존 Python 알고리즘은 모든 시나리오를 미리 코딩해야 하지만, AI 에이전트는 예외 상황을 스스로 인식하고 대응할 수 있다고 주장한다. 최근 리밸런싱(10월 21일): 기존 자산 바구니(SHV, UPRO, UDAO, TQQQ)를 유지하기로 했다. 성장 분석가는 나스닥 급락 이후 UPRO와 TQQQ로 옮기는 것이 합리적이라고 판단했다. 인플레이션 분석가는 유가, 국채 인버스 등 대안을 검토했으나 확인된 충격이 없어 채택하지 않았다. 감사 에이전트는 현재 레버리지가 이미 목표 비중에 가깝다고 판단했고, 트레이더도 레버리지 확대 없이 보유에 동의했다. 진행자는 AI의 사고 과정을 들여다볼 수 있다는 점을 흥미롭게 소개했으나, 개별 종목 가격 전망이나 목표가는 제시하지 않았다.
다듬은 글
This is not Python code. These are not if statements. These are AIs that communicate with each other and argue about what they are doing. As of today, that score is at 111,000, up almost 112, while the S&P has only risen to 105. This is in the midst of a huge market crash, and it is still holding. He went through it all anyway .
Why does this AI now possess what it possesses? Five AI agents. We have three of them analyzing different aspects. This is done in order to copy Ray Dalio and his trading style. Truth? So, he is a kind of macro investor. He pays a lot of attention to growth, inflation, as well as debt and liquidity.
These are the three things he pays the most attention to. And he's a big fan of them arguing with each other. So, that's basically what we did here. We created an AI team. There is a growth agent who goes and analyzes growth. Here's an inflation agent who goes and analyzes this. And then the debt and liquidity agent analyzes it.
Each of them takes their own evidence. They bring them together, and then they have an agent of dissent who essentially says, "Who should we listen to?" And then we have another AI agent that actually executes the trades and makes sure we're managing risk properly. This is all done automatically . This is not Python code.
These are not if statements. These are AIs that communicate with each other and argue about what they are doing. You can see here, this is Ray Dalio's strategy against the S&P 500. The dotted line is the S&P 500. You can see here, we're completely ahead of him. Um , both started at about $100,000 on July 29th.
Today, that score is at 111,000, almost 112, while the S&P only rose to 105. So, that's almost triple the return of the S&P 500 over the same time period . Pretty crazy, right? Um, you can see here that since launch it has almost always been above the S&P 500 throughout . And notice, by the way, this is in the midst of a huge market crash, and he's still surviving and has come through it all .
If I click here, I'll go to AI editing. Ahem. I'll say, "Please tell me why this AI has what it has now?" What does it have now, why does it have it, what did the AI argue about and why did they come to this conclusion? Isn't that right? So, I can ask him about it. And he will get to work. The AI will figure it out and say, "Well, what's going on here?
This is an AI-based strategy. Let's figure out why they're making these decisions? Why are they trading? Why are they, you know, why did they buy this? Why did they buy that?" Isn't that right? You can actually communicate with him, these AI-based strategies are incredibly powerful. Not only can you act like you're hiring an investment team that does all the news research and all the complicated stuff, it's also great because you can do other, more advanced tasks.
Here, Yemi, for example, remember how we used to do all these strategies in Python? Creating such strategies in Python is quite difficult. You have to go through all these little things , like, "Okay, what if this happens? What if that happens?" We actually noticed that the AI-based strategy works much better than Python, because if you give Python a task, you have to consider all possible options, right?
If you forget something, it simply won't work. Python code is not capable of understanding itself. And the AI agent is capable, right? So, even if you have created your own trading algorithm, and this applies to everyone who creates algorithms using Claude, the first thing it will do is just write a bunch of Python code for you.
The problem is, what if he missed something? What if he missed, well, what if the deal doesn't go through? What happens if the prices are wrong? What if, you know, if this, this, there are so many different scenarios that could happen. And if you don't take this into account in Python, it simply won't work.
But if you entrust it to an AI agent, it will know. He'll say, "Hey, I know this situation. This doesn't look right." He'll figure it out and do it anyway. Isn't that right ? So, you can use very complex strategies with agent trading, as it is now called. This is very popular on Robinhood and similar services.
But you can do very advanced things very easily with this kind of technique. Okay , here's what I bought. And here's what we have now. We own SHV, UPRO, UDAO, TQQQ, got it? So right now the portfolio is pretty heavily overweighted with stocks. Okay, and during the last launch on October 21st, the team decided to leave the existing basket of assets.
No warrants were filed. Okay, but I know there were auctions before, for example on October 12th. Well, it looks like a lot of this was sold . You can actually see all the transactions that have been made. And here is the justification for this. You can literally see what these AI bots were thinking, right?
They actually think for themselves. So, the growth researcher said, "UPRO looks like a more balanced core of the market with 3X leverage. TQQQ offered growth exposure but had a sharper recent pullback and higher volatility. UDAO offered another diversified asset with 3X leverage." So our AIs recognized that the market fell last week.
In particular, the NASDAQ fell last week. So he said, "Hey, I'm getting out of this." Yes? They did it automatically for me and moved me to UPRO and TQQQ. Perfectly. Okay, and inflation researchers have considered alternative conditions, such as oil or Treasury bond inverse funds, and less leverage if conditions worsen, but queries to FRED data have failed, and the limited news has not confirmed any inflation or liquidity shocks.
They view these scenarios as unconfirmed, not as a basis for trading. Then an independent auditor and trader: the auditor found that the current leverage level was already close to a 50-20-25 or 50-25-25 ratio, and the historical comparison did not make rebalancing advisable. The trader agreed with the recommendation to hold the position instead of increasing leverage.
Yes? This is very cool. You are literally peering into the minds of these AIs. What are they doing? Why did they do this?
Comments 0