With the general availability of OpenAI's Deep Research:
https://openai.com/index/introducing-deep-research/

What prompts have you used so far related to finance/trading? It has a lot of potential for those of us who like to do our homework.

Mike

Big Mike 

With the general availability of OpenAI's Deep Research: https://openai.com/index/introducing-deep-research/ What prompts have you used so far related to finance/trading? It has a lot of potential for those of us who like to do our homework. Mike

Funny thing, I just used it today for the first time!

My query was : "I want to investigate the effectiveness of pulling and stacking in futures trading, along with its strengths and weaknesses and how to remedy those weaknesses, considering one has MBO data." - Im no prompt engineer, I will admit that, so if somebody sees an error in how I ask, by all means tell me.

I have used it to create risk management models for my futures trading. I have also used it to create NT 8 indicators for me as well.

I have been experimenting for about 3-4 months with a number of the free LLM. I have been using them mostly for coding help (primarily MQL5 but also Easylanguage). I have found that you need at least a basic understanding of how an indicator or EA code is setup in your trading app to effectively prompt the LLM for help. There is a practical limit to the number of lines of code you can ask them to look at also. The code outputs from the models are generally spectacularly good. When it doesn't work initially, keep going back and it will get there. I have seen a huge improvement in outputs just over the past couple of months in terms of code errors.

I have also experimented to a lesser degree with prompting regarding advice on specific stocks. Again, quite spectacular results, particularly with Grok3.

I have tried ChatGPT, Copilot (which I think is mostly ChatGPT) and lately Grok3, which is nothing short of spectacular in anything you ask it. Not tried DeepSeek to any meaningful level.

One thing though, you have to be careful in how you prompt these models, although less so with Grok. What you don't tell it specifically, it will guess what you are trying to ask, usually correctly. It will also prompt you and refine your query if you respond to its questions.

Give it a try

Big Mike 

With the general availability of OpenAI's Deep Research: https://openai.com/index/introducing-deep-research/ What prompts have you used so far related to finance/trading? It has a lot of potential for those of us who like to do our homework. Mike

Kevin Davey (@kevinkdog) just did a couple of Webinars that used AI (ChatGPT in his case). Definitely has some uses but also definitely not fully there yet. (ie code like Low < Lowest(Low, X)

SMCJB 

Mon Feb 24 @ 1 PM Eastern - "Hey ChatGPT, Give Me Some Unique Algo Entries" Wed Feb 26 @ 1 PM Eastern - "AI Assisted Algo Entry Filters"

Morbec 

I have found that you need at least a basic understanding of how an indicator or EA code is setup in your trading app to effectively prompt the LLM for help.

Agree. As mentioned above you needed to know that the Low can never be lower than the Lowest(Low) and the code should be equal to.

Morbec 

The code outputs from the models are generally spectacularly good. When it doesn't work initially, keep going back and it will get there. I have seen a huge improvement in outputs just over the past couple of months in terms of code errors.

Kind of disagree here. I think it makes lots of basic errors, and when you correct it, it often acknowledges the error and then regurgitates the same thing back to you, with the same error again.

I have been using AI for research, summarizing others information and finding nuggets of goodness in documentation.

Have you ever experienced someone trying to explain something and you know there is something there, but what they have written is impossible to understand. Hope no one takes offense to this but, Al Brooks comes to mind and I really do like his information.

I found the summarization works very well.

Imagine recording someone, say a professor in a college class, getting the transcript, processing it through a summarizer, wala you have class notes. What an educational game changer.

I then tried to teach it to do a simple Higher Low and Lower High study.

The system I used simply could not handle storing the extreme of previous bars to do a future comparison.

But what really was the main issue it faced was the time it took and I my ultimate goal is to do this processing in real time.

I started by providing just a few bars of data (bar number, date, time, high, low) and it sort of worked. I then gave it 200 bars. When it is "thinking" the JavaScript spinner appears. After a few minutes of watching the spinner I cancelled it.

Further more, my eyes could "see" these patterns far faster than the reasoning engine could process it. And this was not even connected to a trading platform.

Yes, absolutely the training I provided is related to the issue. The more exposure to this the more it will "learn".

AI is already working for me from a research of text perspective - an invaluable tool. It sits in my tool box along side my screen capture tool and screen makeup tool. The summarizer is a great way to, well, summarize.

Is there a way to get all the data from one of these threads? It would be super cool to have a summarize button available for threads.

AI has a potential in the very new future for real time processing of a lot of data. Likely the non-free versions are doing this already, but they are out of my price range. The free stuff's ability to work large math problems and more complex programming is still a bit off. I have no doubt this is coming. I am a senior developer and architect and where I work is deeply involved in AI, just ask Matthew McConaughey.

Big Mike 

... What prompts have you used so far related to finance/trading? It has a lot of potential for those of us who like to do our homework. Mike

I created a discussion (chat) in chatGPT to help me create well crafted-prompts which i named "Crafting AI Prompts".

Here is how it goes:
---
Here are six ingredients for crafting a well-written prompt for artificial intelligence tools. I want you to consider them to help me draft my prompts in this conversation. First, I will provide you with the six ingredients, and then you will guide me in constructing a prompt. Once you have understood these six ingredients, ask me if I want to write a prompt on any subject. Okay?
---
chatGPT:

Quoting 

Got it! Go ahead and share the six ingredients, and I'll help you craft precise and effective prompts. Once I understand them, I'll ask if you have a specific prompt you'd like to work on.

---
As an AI assistant, i want you to use these six guidelines or best practices to help me craft well-defined prompts:

1. Assign a Specific Role:
Clearly define the role you want the AI to take on. This helps in providing context and direction. For example, "Act as a marketing manager for a tech startup."

2. Establish Context (Target and Objectives):
Specify the target audience and the objective of the prompt. This ensures that the AI's response is relevant and focused. For example, "Write an article explaining the benefits of AI in healthcare for a general audience."

3. Define a Format:
Specify the structure, length, and format of the response. This could include word count, bullet points, or even visual elements. For example, "Create a 500-word blog post with bullet points and at least two images."

4. Specify the Desired Tone:
Indicate the tone you want the AI to adopt, whether it's formal, friendly, humorous, or educational. For example, "Write a friendly and approachable social media post about our new product launch."

5. Set Clear Limits:
Provide specific guidelines to ensure the content meets your criteria. This could include character limits, sentence length, or any other constraints. For example, "Write a Twitter post in less than 280 characters."

6. Iterate and Refine:
Continuously test and refine your prompts based on the responses you receive. Analyze the results and make adjustments to improve the output. For example, after receiving a response, you might say, "Can you make it more engaging and add a call to action?"
---

Quoting 

Got it! These six guidelines will help create well-structured and effective prompts. Would you like to craft a prompt on a specific topic now? Let me know what you have in mind, and I'll guide you through the process!

---
Can you create a prompt for this topic? create a strategy for trading the CME ES Futures market
---

Quoting 

Sure! Let’s refine the prompt using the six best practices. Here’s a well-structured prompt for your topic: "Act as a professional futures trader specializing in the CME E-mini S&P 500 (ES) market. Develop a detailed trading strategy that includes entry and exit rules, risk management techniques, and ideal market conditions. The strategy should be designed for intraday trading and incorporate volume profile analysis and market structure. Present the strategy in a structured format with clear bullet points, and maintain a professional and analytical tone. Limit the explanation to 500 words." Would you like to adjust anything, such as the trading style (scalping, swing trading) or technical indicators?

Once i am satisfied with the well crafted-prompt, i open a new chat and paste it within or try it with another AI assistant tool.

Here is what chatGPT answered for the above prompt:

Quoting 

Intraday Trading Strategy for CME E-mini S&P 500 (ES) Using Volume Profile & Market Structure 1. Market Conditions & Setup Preferred Trading Hours: Focus on the primary session (9:30 AM – 4:00 PM ET) for high liquidity. Avoid the first 5–10 minutes due to erratic price action. Volume Profile & Key Levels: Identify Value Area High (VAH), Value Area Low (VAL), and Point of Control (POC) from the prior session. Mark key high-volume nodes (HVN) as support/resistance and low-volume nodes (LVN) as rejection areas. Use the developing volume profile for intraday context. Market Bias: Trend Day: Expect continuation if price holds above/below the previous day’s value area. Range-Bound Day: Fade moves at VAH/VAL with confluence from order flow signals. 2. Entry Criteria Breakout Strategy (Trend Day): Bullish Setup: Enter long when price reclaims VAH and holds above with increasing delta. Bearish Setup: Enter short when price breaks below VAL and holds with declining delta. Confirmation: Use a footprint chart to validate absorption/exhaustion. Reversion Strategy (Range Day): Long Setup: Buy near VAL if price rejects it with a bullish order flow signal (e.g., excess wick, bid absorption). Short Setup: Sell near VAH if price rejects it with bearish order flow (e.g., offer absorption, trapped buyers). Failure Point: Stop out if price closes beyond value area extremes. 3. Exit Rules Profit Targets: First target: POC or mid-range. Second target: Opposite value area extreme (VAH/VAL). Extended target: Next HVN or previous session’s high/low. Stop Loss Placement: Breakout Trades: Below VAL (for longs) or above VAH (for shorts). Reversion Trades: Beyond the rejection wick or next key level. Max Risk per Trade: 0.5% of account size or 8–12 ticks. Trailing Stop: Move stop to breakeven after price reaches the first target. Use developing VWAP or dynamic delta shifts to adjust stops. 4. Risk Management Daily Loss Limit: Stop trading after three consecutive losses or exceeding 2% of the account. Trade Size: Adjust contract size based on volatility (e.g., lower size during high ATR periods). Correlation Check: Avoid overleveraging when correlated markets (NQ, RTY) signal divergence. News Awareness: Reduce position sizing before major economic releases (FOMC, CPI, NFP). 5. Execution & Review Trade Execution: Enter trades via limit orders at key levels to minimize slippage. Use market orders only for momentum-based breakouts. Post-Trade Analysis: Review executed trades with footprint charts and volume distribution. Log trade rationale, risk/reward, and emotional state. By combining volume profile analysis with market structure, this strategy provides clear trade setups with well-defined risk controls, maximizing intraday trading efficiency in the E-mini S&P 500 market.

Eubie 

Funny thing, I just used it today for the first time! My query was : "I want to investigate the effectiveness of pulling and stacking in futures trading, along with its strengths and weaknesses and how to remedy those weaknesses, considering one has MBO data." - Im no prompt engineer, I will admit that, so if somebody sees an error in how I ask, by all means tell me.

Here is how chatGPT reformulated your prompt using the guidelines i posted above and see how the answer differs from your original prompt:

Quoting 

Act as a professional futures trader with expertise in market microstructure and order flow analysis. Investigate the effectiveness of pulling and stacking in futures trading, focusing on how these order book dynamics influence price action. Analyze the strengths and weaknesses of using pulling and stacking as a trading signal, and propose ways to mitigate the weaknesses. Consider that the trader has access to MBO (Market by Order) data. Structure the response with clear sections, including definitions, analysis, practical applications, and solutions to weaknesses. Maintain a professional and data-driven tone, limiting the explanation to 600 words.

SMCJB 

Kevin Davey (@kevinkdog) just did a couple of Webinars that used AI (ChatGPT in his case). Definitely has some uses but also definitely not fully there yet. (ie code like Low < Lowest(Low, X) Agree. As mentioned above you needed to know that the Low can never be lower than the Lowest(Low) and the code should be equal to. Kind of disagree here. I think it makes lots of basic errors, and when you correct it, it often acknowledges the error and then regurgitates the same thing back to you, with the same error again.

I found the same thing initially, particularly with ChatGPT. For example, it would mix up MQL5 functions with MQL4. But I have seen a huge improvement just in the past couple of months. I get hardly any errors with Grok3 and when I do, it acknowledges and fixes it usually with the next prompt. At least with MQL5 anyway.

I am also finding my coding improving as I learn from it.

Morbec 

I found the same thing initially, particularly with ChatGPT. For example, it would mix up MQL5 functions with MQL4. But I have seen a huge improvement just in the past couple of months. I get hardly any errors with Grok3 and when I do, it acknowledges and fixes it usually with the next prompt. At least with MQL5 anyway.

I wonder if it's an EasyLanguage thing. I find that their VBA code is excellent. But when it comes to Easylanguage they use outdated functions, try to assign variable names that are reserved words etc etc. At this point I've only tried ChatGPT and Perplexity. I was paying for the better version at one point but decided the results were questionable enough that it wasn't worth it.