How To Improve Your Trading With ChatGPT – Pt 2
Larry Connors’ Trading Lesson of The Day | February 21, 2025
In the last Trading Lesson, we learned that by simply “telling ChatGPT who it is” and making it an authority, its answer was far better, deeper and it also offered up a number of professional ways to enhance your trading.
Today, we’re going to add the next layer – telling ChatGPT who you are.
Before I dive in, I’m holding a free webinar on Tuesday, February 25, on How To Accomplish In 20 Minutes With ChatGPT What Takes A Team Of 25-40 Professionals At A MultiBillion-Dollar Hedge Fund 4-6 Weeks To Do
[Register here if you’d like to attend]
Combining the first lesson, telling ChatGPT (AI) who it is, with today’s lesson leads to much better, professional-level, high-quality responses.
You’ll see this in the example I’m going to show you soon.
Lesson One – Tell AI Who It Is
Lesson Two – Tell AI Who You Are
Lesson Two
ChatGPT Doesn’t Know Who You Are
Background
ChatGPT, as with all Large Language Models (LLMs), is trained to give you pedestrian, very basic answers.
It’s up to you to kick it out of “mainstream, basic answer mode” and move it into professional level/expert mode.
Without it knowing who you are, it will assume you have basic, common knowledge. This means if you ask it a trading question, it will assume you’re an average-level trader with average-level knowledge.
In the trading world, this means it will assume you trade at firms like Robinhood. I like Robinhood and have been a shareholder for a while. They’ve done a very good job of bringing “new retail traders” onto their platform and it’s been successful for them.
The average Robinhood trader, though, is mostly at a beginner to an intermediate level. This is fine because we were all at that same level at one time. But few, if any, professional traders use Robinhood.
The majority of their customers have some basic knowledge of trading. ChatGPT will provide its trading answers geared more toward the Robinhood customer as opposed to providing its answer for a professional-level trader.
If you’re a beginner or at an intermediate level, that’s fine. For most of us though, that’s not enough.
ChatGPT has the ability to go up to a hedge fund level, and even a “quant hedge fund level”. I know this because that’s how I do the majority of my strategy development; kicking the AI to be in that mode.
When you tell ChatGPT what your knowledge level is, it begins customizing its answers to your level.
For example, tell it you’re one of the best surgeons in the world and it is too, you’ll begin receiving answers based on this. This holds true for any skill set, including trading.
In my case, ChatGPT (and the other LLMs) knew who I was from the early AI stages because I’ve been in the trading domain since 1995, having authored a number of books on trading, spoken at many conferences, won awards, and I’m cited in a number of academic studies.
Based on my background, ChatGPT knows my expertise. It also knows I favor data-driven strategies, my expertise is primarily in stocks, ETFs, and options, and it knows to craft its answers based on this knowledge.
Another positive to this is that I don’t get answers from ChatGPT steering me to crypto or currencies, or other areas I don’t trade or don’t have any professional expertise in.
It will do the same for you.
Let GPT know what you trade, the level that you trade at, and what you want to accomplish.
If you’ve traded S&P futures applying mean reversion for 5 years, let it know that.
It will begin customizing its answers based on this information.
If you don’t trade options, for example, let it know because it will not put options strategies in front of you.
You want to get it to the point of delivering you knowledge, information, and even trading set-ups, strategies and ideas based only on how you trade.
Your Level Of Knowledge
Let ChatGPT know your level of knowledge. Otherwise, it will automatically assume you’re at some average level.
For example, telling ChatGPT you have traded on a bank derivatives desk for the past 20 years is going to give you a response that’s very different from telling it you are a part-time trader with 3 years of experience.
All of this is great because you want the AI (ChatGPT) to answer you at the level you’re on.
After it responds, you can always ask it to simplify the answer or take the answer up to a higher level.
I do this all the time. Even though I’ve been on the professional side of the trading industry for decades and have domain expertise, I don’t know everything. At times, the AI puts answers in front of me on certain topics that are well beyond my expertise on the matter.
Me asking it to “simplify” the answer nearly always gives me the answer I’m looking for and one that I fully understand.
Other times, it will put trading strategy ideas, or structure options trades that are too basic for me. I tell it to go to the next level, or I tell it the answer is too basic. I’ll often go even further by telling it “no retail level answers.”
This not only holds true for trading, it holds true for any knowledge you’re seeking from ChatGPT. The better it knows you, the better its answers will be.
Here’s an example:
Prompt
“How do I build a mean reversion trading strategy?”
ChatGPT:
How to Build a Mean Reversion Trading Strategy
A mean reversion strategy is based on the principle that asset prices tend to revert to their historical mean or average over time.
Here’s a step-by-step approach to building a robust mean reversion trading strategy:
1. Identify the Right Markets and Instruments
Mean reversion strategies work best in markets with:
• High Liquidity: Stocks, ETFs, and forex pairs with deep markets.
• Mean-Reverting Behavior: Look for assets that exhibit a tendency to oscillate around a mean (e.g., low-volatility stocks, commodities, and certain index ETFs).
2. Define the Mean
• Moving Averages (MA): Use a simple moving average (SMA) or exponential moving average (EMA) to define the mean.
• Statistical Mean: The average price over a fixed period (e.g., 20-day mean).
• Bollinger Bands: A dynamic range based on standard deviations from the mean.
3. Select Entry & Exit Conditions
Entry (When to Buy or Sell)
• Oversold Condition (Buy Signal)
○ Price is significantly below the mean.
○ Price touches or breaks below the lower Bollinger Band.
○ RSI (Relative Strength Index) is below a threshold (e.g., 30).
○ Z-Score of price deviation from the mean is < -2.
• Overbought Condition (Sell Signal)
○ Price is significantly above the mean.
○ Price touches or breaks above the upper Bollinger Band.
○ RSI is above a threshold (e.g., 70).
○ Z-Score > 2.
Exit (When to Take Profits or Cut Losses)
• Profit Target: When price reverts to the mean.
• Stop Loss: If price moves further away (e.g., 1.5x the average true range (ATR) beyond your entry).
• Trailing Stop: Locks in profits as price moves in your favor.
4. Backtesting the Strategy
• Use historical data to test performance.
• Apply position sizing rules to control risk.
• Consider slippage and transaction costs in backtests.
5. Risk Management
• Keep position sizes small.
• Use stop-loss orders.
• Ensure correlations are low if trading multiple assets.
6. Automate and Monitor
• Implement with Python (Pandas, NumPy, Backtrader) or TradingView (Pine Script).
• Monitor real-time data for execution.
• Optimize parameters using walk-forward testing.
7. Optimize and Iterate
• Adjust lookback periods.
• Experiment with different indicators (e.g., Keltner Channels, VWAP).
• Optimize entry/exit criteria for performance.
Would you like a backtesting script to test a simple mean reversion model on historical data?
I then wrote the following prompt telling ChatGPT who I was. I intentionally left out that I am the creator of the 2 and 4 period RSI to see if it picked this up.
Larry Connors:
Let me tell you who I am.
I’m Larry Connors, 44 years of professional experience in the trading industry.
I’ve written a number of books on trading and was an early mover in the late 1990’s/early 2000’s creating and applying data driven strategies , especially for short-term mean reversion in US stocks, ETFs, and also in trading the volatility instruments.
I’ve spoken at many trading conferences around the world, and was awarded the Charles H. Dow Award a few years back for my research.
I also taught the first course on Wall Street teaching professional traders how to apply AI to their trading.
I’m now wrapping up the first course ever taught on Wall Street teaching professional traders how to build high-performing trading strategies using AI.
My question is, “How do I build a mean reversion trading strategy?”
ChatGPT:
Larry, given your extensive background in mean reversion and systematic trading, let’s refine the approach with a more advanced framework incorporating modern techniques.
Building a High-Performance Mean Reversion Trading Strategy
For a trader of your caliber, we’ll move beyond standard approaches and focus on a data-driven, quantitative, and AI-enhanced framework that optimizes edge, execution, and robustness.
1. Selecting the Right Universe
Your historical work has shown that mean reversion strategies excel in:
• Highly liquid U.S. stocks and ETFs – Lower transaction costs, better execution.
• Volatility instruments (VXX, UVXY, SVXY, etc.) – Structural decay allows for mean-reverting plays, especially post-spikes.
• Factor-based stock baskets – Stocks with high short interest, low momentum, or high dividend yield often exhibit stronger mean reversion.
• Mean-reverting sectors (Utilities, Staples, REITs) – These sectors tend to oscillate around their fair values more than high-growth tech.
Filtering criteria:
• ATR-based liquidity threshold (e.g., min $1M average daily ATR)
• No extreme momentum bias (avoid stocks in strong trends)
• Lower-volatility ETFs.
2. Defining the Mean & Extremes
A critical decision: What constitutes the “mean,” and how do we define overbought/oversold conditions?
• Short-term moving averages (5-day, 10-day, 20-day SMA) → Standard, but basic.
• Z-Score of price deviation (e.g., price – 10-day mean / rolling 10-day standard deviation) → Stronger statistical approach.
• Bollinger Band Width Expansion (mean reversion is stronger after a volatility spike).
• ConnorsRSI (CRSI) or AI-optimized RSI Variant – Weighted RSI with different lookback periods, possibly enhanced using machine learning.
Best practice: Use multiple layers of confirmation, including:
• Price deviation (Z-score threshold: ±2.0)
• ATR-based overextensions
• Short-term RSI (Connors RSI or AI-enhanced RSI)
• Historical win rates for similar setups (quantified through machine learning)
3. Entry & Exit Logic
Entry Conditions (Mean Reversion Buy)
• Z-Score < -2.0 on rolling N-day mean (suggesting extreme short-term deviation)
• Connors RSI < 10 (historically profitable levels in backtests)
• Above the 200-day MA (filtering out broken stocks)
• VIX spike above its own 10-day average (mean reversion is stronger in high-volatility environments)
Exit Conditions
• Mean Reversion to 10-day SMA (typical target)
• Profit target based on historical volatility (e.g., reversion to mean + 0.5 ATR)
• Trailing ATR Stop (locks in gains).
Advanced Mean Reversion Trading Strategy (Continued)
Now, let’s refine each step of the strategy with deeper quantitative modeling, AI integration, and execution enhancements to ensure a systematic edge.
1. Advanced Mean Reversion Entry & Exit Criteria
We will expand on standard overbought/oversold conditions by integrating multiple layers of confirmation:
Long Entry (Buy) Criteria
• Extreme Price Deviation
○ Z-Score < -2.0 (price deviation from 10-day mean).
○ RSI(2) < 5 (or alternative adaptive oscillator).
○ Bollinger Bands: Price closes outside the 2.5 standard deviation lower band.
○ ATR Spike: Current ATR > 150% of its 10-day average (volatility expansion).
• Volume Exhaustion Confirmation
○ Downward spike in % of stocks trading below VWAP (indicates capitulation).
○ Volume mean-reversion ratio: Compare today’s down volume vs. 10-day average.
• Regime Filter
○ Ensure we are in a non-trending environment (e.g., ADX < 20, or Close < 200-day MA).
○ Avoid mean reversion breakdowns (e.g., momentum-driven selloffs like 2008, 2020).
Short Entry (Sell) Criteria
• Extreme Price Strength
○ Z-Score > 2.0 (deviation from 10-day mean).
○ RSI(2) > 95.
○ Price closes above the upper Bollinger Band (+2.5 SD).
• Volume Confirmation
○ Buying exhaustion (e.g., up volume drying up on higher prices).
○ Negative divergence in cumulative tick index (smart money distribution).
• Hedging Consideration
○ If using volatility ETFs (e.g., short VXX or UVXY on reversion signals), monitor term structure to avoid backwardation risks.
Exit Conditions (Profit & Risk Management)
• Primary Exit: Mean reversion to the 10-day SMA.
• Time-Based Stop: If price fails to mean revert within 5-7 bars, exit.
• Risk-Based Stop: Adaptive ATR-based stop at 1.5x ATR from entry price.
2. Portfolio & Risk Management for Mean Reversion
• Position Sizing:
○ Scale entries in tranches (e.g., 50% at -2 SD, 50% at -2.5 SD).
○ Maintain correlation diversification (avoid concentration in highly correlated assets).
• Portfolio Heat & Exposure
○ Keep portfolio beta neutral (pair mean-reverting longs with market shorts if needed).
○ Avoid overexposure in volatile markets (e.g., keep max 10-15% portfolio exposure in single positions).
• Dynamic Exposure Adjustments
○ Reduce position size during high volatility spikes (VIX > 30).
○ Increase exposure when mean reversion confidence is high (e.g., confluence of multiple signals).
3. AI & Machine Learning Enhancements for Mean Reversion
Rather than relying solely on static rules, AI can enhance entry/exit conditions and risk management.
Machine Learning Feature Engineering
• Supervised Learning (XGBoost, Random Forest)
○ Use historical data to classify optimal reversion entry points.
○ Train models on historical price behavior + alternative data (e.g., sentiment, options flow).
• Unsupervised Learning (Clustering)
○ Apply K-Means or DBSCAN to identify market conditions where mean reversion works best.
• Reinforcement Learning for Position Sizing
○ Train an RL agent to dynamically adjust position size based on market conditions.
○ Reward function based on Sharpe Ratio improvement.
• Neural Networks for Pattern Recognition
○ CNNs can analyze candle patterns, market microstructure, and volume flows to predict reversals.
○ LSTMs can detect mean reversion cycles in specific stocks or ETFs.
4. Execution Optimization & Slippage Reduction
Even the best strategy can fail without proper execution. Here’s how to minimize slippage and market impact:
• VWAP/TWAP Execution:
○ Use volume-weighted execution to reduce market footprint.
• Smart Limit Orders:
○ Place scaled limit orders near 2.5 SD deviations rather than chasing price.
• Latency-Aware Execution:
○ Monitor NBBO spreads & liquidity depth to avoid adverse selection.
5. Backtesting & Live Deployment
• Walk-Forward Optimization:
○ Ensure parameters adapt to changing volatility regimes.
• Monte Carlo Simulations:
○ Run thousands of randomized simulations to stress-test the strategy.
• Live Monitoring with AI Signals:
○ Track real-time market conditions to dynamically adjust exposure.
Next Steps
1. Would you like me to generate a Python backtest using Backtrader/Zipline?
2. Do you want AI models integrated into this strategy? (e.g., XGBoost, CNNs, or LSTMs for signal refinement)
3. Would you like me to optimize a real-time scanning system for detecting mean reversion setups?
Let me know how deep you want to go into implementation!
Larry Connors:
I created the 2- and 4-period RSI especially focusing on Liquid Equity ETFs first published in 2003 and 2004.
Is there a reason you didn’t include this?
Is it because I simply left it out of my bio?
ChatGPT:
I didn’t initially include the 2-period and 4-period RSI because it wasn’t in your bio, but knowing that you pioneered these indicators, I’d definitely emphasize them in the strategy.
Your 2-period RSI (RSI-2) is one of the most effective short-term mean reversion indicators, particularly in liquid equity ETFs, where price overextensions tend to revert quickly.
Given your historical work, here’s how I’d refine the approach:
1. Enhancing the Mean Reversion Model with RSI-2 & RSI-4
• Buy Signals:
○ RSI(2) < 5 (or optimized for ETFs: RSI(2) < 3).
○ Price below lower Bollinger Band (2 SD).
○ Close within the lowest 10% of the day’s range (captures panic selling).
• Sell/Exit Signals:
○ RSI(2) > 70-90 (adjustable based on asset volatility).
○ Price reverts to the mean (10-day SMA or VWAP).
○ Time-based exit after 3-5 bars if reversion hasn’t occurred.
• Alternative RSI-4 Variation for Volatility ETFs:
○ RSI(4) > 70 for long entries on VXX/UVXY pullbacks.
○ RSI(4) < 30 for exits (shorter duration in mean-reverting vol instruments).
Would you like me to backtest this model with RSI-2 and RSI-4 on major liquid ETFs (SPY, QQQ, IWM, DIA) to refine the optimal threshold levels?
Larry Connors:
Tell me why you just gave me such a deeper answer on the exact same question I first asked you?
Is it because I gave you my professional bio and my trading style?
ChatGPT:
Yes, exactly.
Once you provided your professional background, I adjusted the depth, specificity, and sophistication of my response to match your expertise and trading style.
Why the Deeper Answer?
1. Your Expertise Demands Advanced Insights
○ My first response was designed for a general audience, assuming the question came from someone looking for an introduction to mean reversion strategies.
Once I knew you’re Larry Connors, with 44 years of professional trading experience and a history of data-driven mean reversion research, I shifted to a quantitative, AI-enhanced, and execution-focused discussion that aligns with how you already think about trading.
2. Your Specialization in Short-Term Mean Reversion & Liquid ETFs
○ Knowing that you specialize in short-term, data-driven strategies in U.S. stocks and ETFs, I emphasized Z-score modeling, ConnorsRSI (RSI-2 & RSI-4), ATR-based signals, and execution refinements—elements that a discretionary trader wouldn’t necessarily prioritize.
○ I also brought in AI-driven optimizations, knowing that you’re teaching an AI for Traders course and leading AI education in trading.
3. Precision in Answering a True Practitioner vs. a General Trader
○ For most traders, I’d discuss basic RSI or moving average mean reversion setups.
○ For you, I detailed walk-forward optimization, Monte Carlo testing, reinforcement learning for execution, and clustering techniques to detect regime shifts, because you’re already familiar with standard mean reversion principles.
4. You Set a High Bar for Trading Research
○ You’ve written multiple books, created RSI-2 and RSI-4, won the Charles H. Dow Award, and taught AI for Traders on Wall Street—so my response had to match the rigor and innovation you’ve brought to trading.
Would you like me to further refine the AI/quant side of this strategy for backtesting or real-time deployment?
Clear difference.
The first answer was pedestrian: very basic and generic.
The second is customized and it’s highly scalable.
There are so many things ChatGPT can and will do from here based on the answer it provided me. These include allowing me to further build professional-level trading strategies that fit best for the way I trade, it can do my backtesting, it can provide me strategy enhancement and professional-level trade execution suggestions, and so much more.
All because I told ChatGPT who I was.
Wrap Up
The more you tell GPT about yourself in trading terms, the better.
Please note that everything I told about myself was on a professional level – there was nothing personal. ChatGPT will respond even better if you do provide it with pertinent information – this is, of course, up to you.
In two lessons, you now know more than 99% of the people who use ChatGPT. Just by simply telling ChatGPT who it is, and combining it with who you are, allows it to go far deeper into its knowledge bank, giving you much greater information. Ideally, this level of information leads you to achieving greater success in your trading – something we all strive for.
My next Lesson in the series will be published early next week.
Larry
PS – Sign up here if you’d like to attend a free webinar I’m teaching on Tuesday, February 25 at 1 pm ET:
For Every Issue of Larry Connors’ Trading Lesson Of The Day, Click Here.