GPT 통합을 통한 트레이딩 전략 자동화

### 자동매매의 발전: 향상된 의사결정을 위한 GPT 통합
* 📈 이 영상에서는 타코즈랩 어카운팅의 윤상철 님이 **파이썬과 한국투자증권 API를 이용해 개발한 자동매매 프로그램 개발 과정**을 공유합니다.
* 🤖 그는 알고리즘을 개선하기 위해 **새로운 아이디어를 테스트**하여 수익성을 높이고 개인 투자 스타일에 맞게 트레이딩 전략을 조정했습니다.
* 📊 윤은 트레이딩 프로그램이 시장 데이터를 전송하고 실시간 안내를 받을 수 있는 **GPT를 투자 의사 결정에 활용하는 방법**에 대해 설명합니다.
* 📉 그는 **GPT가 3년간의 주식 시장 데이터를 분석**하고 MACD와 RSI와 같은 기술적 분석 방법을 기반으로 투자를 추천하는 실기 테스트를 강조합니다.
* 💡 윤은 **GPT와 매일 상호 작용**하여 주식 추천 및 거래 지시를 받고, **10분마다 매수, 매도 또는 보유 결정**을 자동화하는 것을 구상하고 있습니다.
* 💳 그는 자신의 트레이딩 시스템에 통합하여 **실시간 쿼리 및 응답이 가능한 OpenAI API의 비용 효율성**에 대해 언급합니다.
* 📊 향후 계획에는 **실시간 뉴스 업데이트를 통합**하고 **개인 투자 데이터베이스를 GPT에 연결**하여 보다 정확한 거래 결정을 내릴 수 있도록 하는 것이 포함됩니다.
* 🔍 전반적으로 윤 대표는 트레이딩 전략 자동화에 있어 **GPT의 잠재력**에 대해 기대감을 표하며, 다음 동영상에서 **추가 개발 사항**을 공유하겠다고 약속했습니다.

윤상철 트레이더는 투자 의사 결정과 효율성을 높이기 위해 자동화된 트레이딩 전략에 GPT를 혁신적으로 통합했습니다.

Hello, this is Sangchul Yoon from TacozLabs Accounting. In this video, I want to share the advancements I've made in the automatic trading program I've been working on. I've tested a new idea to enhance the algorithm, and I'm excited to share this progress with you. In the next sessions, I plan to implement and test this approach further. As you might have seen in the thumbnail, up to now, I've been developing an automatic trading program using Python, leveraging the user-friendly API provided by Korea Investment & Securities. I've shown you how this setup can facilitate automatic trading. However, to improve profitability and tailor the trading to my preferences, I've realized the algorithm needs further refinement. This led me to consider the various technical analysis techniques available for day trading and how they could be adapted to the Korean stock market or my investment style. Most of you are likely working full-time and understand how challenging it is to find time

for such projects after work. This is where I thought of utilizing GPT, which has become quite accessible in recent years. I've subscribed to the OpenAI API and have been using it for tasks such as holding meetings with international clients or composing emails in English. It occurred to me that GPT could also assist in investment decision-making by enhancing the algorithm for my trading program. The idea was to use GPT for making investment decisions, which could then be communicated in real-time to the trading program. The program would send market data to GPT, which would return decision-making guidance. To implement this, I considered how to design the system. I chose to leverage the GPT API, which has been available for about a year now. Typically, we interact with GPT through a web interface by posing questions and receiving answers. However, I wanted to explore a more engaging application. For example, in one of my tests, I used publicly available libraries to download stock market

data. I uploaded a file named GPT_data_all.csv, which contained three years of price data (opening, closing, highest, lowest prices, and volume) for five companies listed on the KOSPI. I then asked GPT to analyze this data from 2021 for day trading purposes, inquiring about applicable technical analysis techniques. GPT listed several methods, such as MACD and RSI. With this information, I queried GPT on which of the five stocks it would recommend investing in immediately based on these technical analysis methods. GPT proceeded to analyze the data for all five tickers using the MACD technique, among others, and provided its recommendation on which stock seemed most promising. This approach allowed GPT to rank the five tickers based on various technical analysis criteria, using the dataset I provided. It's an innovative way to combine AI capabilities with stock market investments, showing the potential of using GPT not just for generating text but for making informed financial decisions.

And then, finally, it's a bit long, but by saying do this, it produced this result. So, there are five tickers, and it used four analytical methods to predict which ones would be effective for investment according to each method, leading to outcomes for each ticker. Then, posing the ultimate question: If you were day trading, what would you do tomorrow? It makes a decision from there, choosing ticker 5490. As you know, our country's ticker symbols are six digits, so if 5490 has only four digits, it implies the missing zeroes at the beginning. It then discusses why investing in this option could be advantageous. This made me think, essentially, I'd be conversing with it daily, asking, Hey, what's happening in the KOSPI, the Korean stock market? Give me the data, what technical analysis items are available? and so on. Then, about an hour before the market opens, tell me the top ten stocks you'd recommend.

buying at the opening. After receiving this, my automated trading program processes these ten pieces of data, performing trades based not only on purchases. It continuously asks GPT throughout the day whether to buy, sell, or hold based on new data, following its instructions for trading. My setup involves sending GPT trade data for the chosen stocks every ten minutes from the opening. GPT is then queried, Based on your advice, I've bought this; here's the current trade data for these ten stocks, what should I do now? So, it's about deciding buy, sell, or hold actions among these ten stocks, every ten minutes, following GPT's instructions. This seemed like a fun approach. While I could manually input technical trading algorithms into the code, which would take time but is feasible, the downside of handling this personally, especially for those not trading full-time, is the consistent lack of time and unpredictable work hours.

I felt relying solely on my manual intervention wasn't wise; hence, the idea of letting GPT take on the role, analyze outcomes, and apply continuous trading strategies based on its analysis. It could also report back to me, making the process more convenient. So, I thought, 'Why not?' I've confirmed that theoretically, it's feasible to connect GPT to automated trading, and GPT can indeed make data-driven decisions based on technical analysis. Next steps? Visit the OpenAI website to explore the API services. The website looks great, and here you can see how to connect to their API services. I've already paid for the service; it's not expensive, just link a card. Testing this, you can track your spending, seeing charges for each use. With about $9 left, I reviewed the pricing policies for using the API with different GPT versions, deciding on GPT-3.5 Turbo for cost-efficiency.

Registering my API key allowed me to link my automated trading with GPT, which I'll demonstrate. Secondly, after receiving the API key from OpenAI, I tried connecting it to see if it accurately responds to queries. You can find detailed instructions online for setting up and using the API for queries. Before sending text prompts to GPT, you can assign it a role to specify how it should respond. I gave it a simple role, You are a highly skilled fund manager investing in Korea, applying famous investment strategies... Asking it to recommend stocks based on this setup, I tried running a test with GPT-4. Let's see how it responds. Expecting initial delays, I awaited its advice based on the given investment strategy, querying for stock recommendations. Though I hadn't provided specific data yet, GPT offered a general description of the strategy and asked for data to proceed. This interaction showcases the potential for utilizing GPT in automating trading decisions.

furthering the integration of AI in personal investment strategies. So, utilizing the GPT API for my automated trading program seems entirely feasible. The key is to refine the prompts meticulously, ensuring they precisely match the dataset required for investment decisions. I plan to send queries every ten minutes, asking whether to buy, hold, or sell, and expect responses in the form of a data table. This table will directly influence the trading decisions made by the program. Thus, I've concluded it's possible to integrate GPT effectively. Instead of crafting algorithms myself—acknowledging my limitations in knowledge and time—I'll rely on GPT, which I've already integrated into my trading setup. This setup will continually receive investment decisions based on predefined strategies, significantly automating the process. My role for GPT will be defined as a day trader aiming for a 1% return with an initial capital of 1 million won, allowing it to freely compose a portfolio.

Over time, I can adjust this portfolio, deciding the percentage to allocate to blue-chip stocks, potential high-gain stocks, etc. In addition to market data, I plan to include real-time news updates by connecting a news API, incorporating breaking news into the decision-making process. Though I haven't tested this yet, it appears to be a feasible goal. Another intriguing aspect I discovered while studying the GPT API is its potential for custom application. For example, I've been collecting data on investments and building a database (DB). I realized that it's possible to link this DB to GPT, allowing it to use my data for more precise responses. This capability extends beyond automated trading to practical applications in report writing, offering detailed numerical data and current business conditions. It opens up possibilities for leveraging my data collection in conjunction with GPT for various purposes. In summary, this video discussed my plans to utilize GPT for automating trading strategies.

The potential for GPT to streamline tasks and reduce the need for manual resource allocation in both professional and hobbyist applications is exciting. I plan to connect everything practically and share the outcomes in future videos. If you have any questions, feel free to comment below, and I'll see you in the next video.

GPT4 가 자동매매 프로그램에 연결되면 어떤일이? What Happens When GPT-4 Connects to an Automated Trading Program

재무·금융 지식과 데이터 분석 tacozlabs.com


야근하는 회계사

자동 트레이딩 프로그램의 발전

타코스랩의 윤상철 회계사는 파이썬과 한국투자증권 API를 활용한 자동매매 프로그램을 개발 중입니다.

수익성을 높이기 위해 알고리즘을 다듬고 데이트레이딩을 위한 기술적 분석 기법도 고려하고 있습니다.

윤 대표는 실시간 투자 의사결정에 OpenAI GPT API를 활용해 시장 데이터와 가이드를 트레이딩 프로그램에 전달할 계획입니다.

데이 트레이딩에 AI 활용하기: 코스피 종목 분석하기

업로드된 CSV 파일에는 코스피 상장사 5곳의 3년치 가격 데이터가 담겨 있었습니다.

GPT는 즉시 투자할 수 있는 최적의 종목을 추천하기 위해 MACD, RSI 등 기술적 분석 방법을 제공했습니다.

분석은 다양한 기준에 따라 시세 순위를 매겼고, 최종적으로 티커 5490을 데이 트레이딩에 가장 유망한 옵션으로 제안했습니다.

트레이딩 프로그램은 데이터를 처리하고 10분마다 10개 종목에 대한 매수, 매도 또는 보유 결정을 위해 GPT에 쿼리합니다.

이 접근 방식은 수동 트레이딩의 시간 제약을 완화하여 GPT가 결과를 분석하고 지속적인 트레이딩 전략을 적용할 수 있도록 하는 것을 목표로 합니다.

5 터보 버전을 선택해 OpenAI의 GPT 서비스용 API에 연결할 계획입니다.

OpenAI에서 API 키를 등록하면 GPT를 자동매매 시스템과 연결할 수 있었습니다.

저는 GPT에 한국에 투자하는 펀드 매니저 역할을 할당하고 특정 투자 전략에 따른 주식 추천을 쿼리했습니다.

매수, 보유, 매도 결정을 위해 10분마다 쿼리를 보내고, GPT의 응답을 데이터 테이블에 담아 트레이딩 프로그램의 결정에 영향을 미치도록 할 계획입니다.

GPT로 트레이딩 전략 자동화하기

포트폴리오는 시간이 지남에 따라 우량주 및 고수익 종목에 초점을 맞춰 조정되며, 뉴스 API를 통해 실시간 뉴스 업데이트를 통합할 예정입니다.

투자 데이터 데이터베이스를 연결하여 보다 정확한 대응과 보고서 작성에 실질적인 적용을 가능하게 함으로써 GPT API의 맞춤형 적용 가능성을 모색하고 있습니다.

목표는 작업을 간소화하고 수동 리소스 할당을 줄이는 것이며, 향후 동영상을 통해 그 결과를 공유할 계획입니다.