Can AI Predict Stock Returns Using Technical Indicators?
Can Technical Indicators Predict Future Returns?
The debate on whether past price and volume data can predict future stock returns is ongoing. Some believe that you can use historical price and volume data to anticipate future returns. This view contradicts the efficient market hypothesis, which argues that current stock prices reflect all available information. However, human behavior isn’t always rational; people suffer from systematic biases which could be reflected in price movements. If true, repetition of these patterns means that historical data can indeed provide insights into future stock movements. Technical indicators are tools used to harness this information.
The QUINETICS Monte Carlo Approach
On QUINETICS individuals can create, test, and trade AI-driven trading strategies with a single click. This feature is powered by a “Strategy Monte Carlo” simulation that continually generates thousands of strategies based on economic, technical, fundamental, and sentiment data for various assets.
By selecting “Load strategy” as depicted below, users can load the AI strategy for a chosen asset and indicator type (in our example, NVIDIA and technical indicators) that demonstrates the highest historical Sharpe Ratio from thousands of AI-generated strategies for this selection.
Monte Carlo Simulation Results for Bollinger Band Strategies
Although the platform focuses on AI strategies, it also allows users to create strategies based on “traditional” technical indicators. Hence, the Monte Carlo simulation can be used to create hundreds of different Bollinger Band strategies across individual US and EU stocks based on the classic interpretation of this indicator: buy if price is below the lower band and sell if the price is above the upper band. Each strategy in the simulation will have varying band widths, moving average lengths, and selling conditions.
We can then look at the results if the traditional Bollinger Band generated any excess returns. The histogram below displays the distribution of Sharpe Ratios resulting from the Monte Carlo simulation for classic Bollinger Band strategies:
As illustrated, the mean Sharpe Ratio is close to zero, with a t-statistic of 1.62, which isn’t statistically significant. It seems that Bollinger Band by itself did not generate any outperformance.
Monte Carlo Simulation Results for AI-Interpreted Bollinger Band Strategies
Now, we repeat the simulation for AI strategies derived from models trained on past Bollinger Band Indicators to determine optimal buy and sell times for an asset based on this indicator. The mean Sharpe Ratio improves dramatically to 0.688 with a t-statistic of 11.85. This is highly significant.
It seems that there is evidence that the AI version of the indicator does a better job than the traditional interpretation in forecasting returns.
Our NVIDIA AI Bollinger Band example
Now, what happened to our NVIDIA example from the beginning? Below, we see NVIDIA’s stock price alongside the Bollinger Bands and buy signals marked in green. The AI predominantly maintained a long position but ceased buying as volatility increased near the stock’s all-time high. Interestingly, the recent market downturn was successfully navigated, with new buy signals at lower prices after the dip.
The backtest reveals a net return of 235% for the AI strategy for the past year, compared to 191% for the stock itself. Also, volatility is reduced with the AI strategy. Of course, backtested performance is no indicator of future performance.
Conclusion
While the comparison between AI and traditional interpretations of technical indicators isn’t necessarily scientifically conclusive due to inherent correlations underlying the Monte Carlo simulation, it still presents a compelling case that AI may interpret historical price movements more effectively.
If you want to see more content like this, follow us on social media:
🏢Follow us on Instagram
🐦Follow us on X/Twitter
🔔Register for free on QUINETICS
Cheers!
Fabian
Visit us at DataDrivenInvestor.com
Subscribe to DDIntel here.
Featured Article:
Join our creator ecosystem here.
DDI Official Telegram Channel: https://t.me/+tafUp6ecEys4YjQ1
