I just tried OpenAI’s updated o1 model. This technology will BREAK Wall Street

When I first tried the o1-preview model, released in mid-September, I was not impressed. Unlike traditional large language models, the o1 family of models do not respond instantly. They “think” about the question and possible solutions, and this process takes forever.

Combined with the extraordinarily high cost of using the model and the lack of basic features (like function-calling), I seldom used the model, even though I’ve shown how to use it to create a market-beating trading strategy.

However, OpenAI just released the newest o1 model. Unlike its predecessor (o1-preview), this new reasoning model has the following upgrades:

• Better accuracy with less reasoning tokens: this new model is smarter and faster, operating at a PhD level of intelligence.
• Vision: Unlike the blind o1-preview model, the new o1 model can actually see with the vision API
• Function-calling: Most importantly, the new model supports function-calling, allowing us to generate syntactically-valid JSON objects in the API

With these new upgrades (particularly function-calling), I decided to see how powerful this new model was.

In the historical example below, I created SPXL rules and compared them with SPY over January 2020 through January 2022. The article reports an outperformance figure and a drawdown near 69%; both matter when assessing the result.

I also performed accurate financial research that even Wall Street would be jealous of.

Let me explain.

Enhanced financial research capabilities

Unlike the strongest traditional language models, the Large Reasoning Models are capable of thinking for as long as necessary to answer a question.

In this recorded request, O1 produced a query that I found useful for the financial question. Its reasoning and the returned query remain inspectable; the predicates, periods and records still need to match what was asked.

For example, I asked the model the following question:

Since Jan 1st 2000, how many times has SPY fallen 5% in a 7-day period? In other words, at time t, how many times has the percent return at time (t + 7 days) been -5% or more.

Note, I’m asking 7 calendar days, not 7 trading days.

In the results, include the data ranges of these drops and show the percent return. Also, format these results in a markdown table.

It gave the following results:

Other language models respond instantly and poorly. Even Claude 3.5 Sonnet, the best traditional language model, failed to give an accurate response to my question.

While it’s possible to tell Claude what it did wrong, it requires reading the query generated by the model and interpreting it.

For example, you have to be a SQL expert to know that this query is wrong because it uses LEAD(..., 7) , which calculates 7 trading days (not 7 calendar days). Additionally, you can see that the model did not respond in markdown format, even though we explicitly asked it to. Then, we have to tell Claude what it did wrong and hope it does it right the next time.

As you can imagine, this is very hard for a non-technical investor to do. And even harder for them to spot the issues with the query.

For the example shown, O1 returned a query on the first attempt without my manually rewriting it. First-attempt generation and verified correctness are different checks.

Transforming these insights into trading strategies

Staying with o1, I had a long conversation with the model. From this conversation, I extracted the following insights:

Essentially I learned that even in the face of large drawdowns, the market tends to recover over the next few months. This includes unprecedented market downturns, like the 2008 financial crisis and the COVID-19 pandemic.

We can transform these insights into algorithmic trading strategies, taking advantage of the fact that the market tends to rebound after a pullback.

For example, I used the LLM to create the following rules:

• Buy 50% of our buying power if we have less than $500 of SPXL positions
• Sell 20% of our portfolio value in SPXL if we haven’t sold in 10000 (an arbitrarily large number) days and our positions are up 10%
• Sell 20% of our portfolio value in SPXL if the SPXL stock price is up 10% from when we last sold it
• Buy 40% of our buying power in SPXL if our SPXL positions are down 12% or more

SPXL’s February 2024 prospectus describes a target of 300% of the S&P 500 Index’s daily performance before fees and expenses. It does not promise three times SPY’s multi-year return. Daily compounding makes the longer-period result depend on the path of returns. These rules add their own buy/sell timing and cash allocation.

I tested the strategy from January 1, 2020 through January 1, 2022. The article reports a maximum drawdown near 69% and an 85% outperformance figure in that test. The prose does not give both period returns or define whether 85% is a relative difference or a percentage-point gap. Read the saved result before making that comparison; the drawdown belongs beside it.

This is just one simple example. In reality, we can iteratively change the parameters to fit certain market conditions, or even create different strategies depending on the current market.

All without writing a single line of code.

Once we’re ready, we can deploy the strategy to the market with the click of a button.

Concluding Thoughts

The OpenAI O1 model is an enormous step forward for finance. It allows anybody to perform highly complex financial research without having to be a SQL expert. The impact of this can’t be understated.

The reality is that these models are getting better and cheaper. The fact that I was able to extract real insights from the market and transform them into automated investing strategies is something that was never heard of even 3 years ago.

The useful workflow is to ask a precise historical question, inspect the generated query, then create explicit rules and compare their saved results. NexusTrade supplies the interface shown here; the result still needs its dates, benchmark and drawdown.

Ready to revolutionize your investing journey? Sign up today and see how NexusTrade can help you take control of your financial future!

NexusTrade - AI-Powered Algorithmic Trading Platform

Learn to conquer the market by deploying no-code algorithmic trading strategies.

nexustrade.io

Follow me: LinkedIn | X (Twitter) | TikTok | Instagram | Newsletter