I Tested 100 AI Indicators So You Don’t Have To (shocking results)

Best AI Trading Indicators: I Tested 100 of Them

I asked Gemini AI to generate 100 different filter conditions for TradeStation. Then I tested every single one across 3,500 breakout strategies inside BreakoutOS.

The result? The best AI trading indicator is not an AI indicator at all. It is a decades-old classic that most traders overlook. Here is exactly how I ran this experiment and what I found.

How I Generated 100 Filter Conditions With AI

Here is what I actually did. I went to Gemini AI and asked it to generate 100 different filter conditions written in EasyLanguage for TradeStation. Each condition was its own “case” with unique input parameters – things like lookback periods, thresholds, and multipliers.

It took about an hour to clean everything up. Gemini gave me duplicates, some nonsensical conditions, and a few that just would not compile. But after cleanup, I had 100 usable filter conditions organized into six categories:

• Moving averages – 15 conditions
• Oscillators – 25 conditions (the largest group)
• Price action – 15 conditions
• Trend – 15 conditions
• Volatility – 15 conditions
• Volume – 15 conditions

That gave me a broad cross-section of everything the trading world has to offer – from simple moving average crossovers to complex oscillator setups that AI loves to generate.

The Testing Framework

This is where it gets serious. I did not just slap these filters onto one strategy and call it a day. That tells you nothing. Instead, I used a rigorous framework built on the Mr. Breakouts Formula.

First, I created thousands of breakout strategies using the formula without any filters applied. These were the baseline. Then I tested each strategy with each of the 100 filter conditions applied on top. That gave me 3,500 total strategy variations imported into BreakoutOS.

The key metric I used is something proprietary to BreakoutOS called the “uplift index.” It measures the overall improvement a filter provides across multiple dimensions at once – net profit, return ratio, win percentage, and average trade size. It is not just asking “did the filter make more money?” It is asking “did the filter make the strategy better in every meaningful way?”

This is critical because most people test one indicator on one strategy and declare victory. That approach is useless. To really know if a filter has value, you need to see consistent improvement across a large sample – the same principle that applies to validating whether a strategy is real.

Cluster Analysis: Which Categories Actually Work?

I ran a cluster analysis across all six categories, looking at both in-sample and out-of-sample performance. The results were brutal for most categories.

Here is what the data showed:

Filter Category	In-Sample Performance	Out-of-Sample Performance
Volume	Positive uplift	Positive uplift
Moving Averages	Positive uplift	Positive uplift
Oscillators (25 conditions)	Negative – made results worse	Mixed
Price Action	Negative – made results worse	Mixed
Trend	Negative – made results worse	Mixed
Volatility	Negative – made results worse	Mixed

Only two categories showed consistent positive uplift both in-sample and out-of-sample: volume and moving averages. Every other category – including oscillators, which had the most conditions at 25 – actually made strategies worse in-sample. That is not marginal underperformance. They actively degraded results.

If you are using RSI, stochastics, or other oscillator-based filters on your breakout strategies, this data should give you serious pause.

The Winning Indicator Is Not AI at All

When I drilled down from categories to individual conditions, one clear winner emerged. It was not a fancy AI-generated oscillator. It was not a machine learning signal. It was conditions 58 and 100 – both measuring the same thing: bar range relative to ATR.

Super simple. Just high minus low, or average true range over some lookback period. That is it.

This indicator has been around for decades. It measures something fundamental about market behavior – how much a bar has expanded or contracted relative to its recent history. And it crushed everything else in the test.

Here is what condition 100 (bar range) delivered when applied as a filter:

• Max drawdown decreased by 40% – that alone makes it worth using
• Net profit improved by 60% – not a typo
• Win percentage improved – more winning trades, fewer losers
• Average trade improved – each trade captured more profit on average

But the most important finding was robustness. Bar range and ATR had the highest robustness index of any filter tested. That means the improvement you see in-sample actually correlates with improvement out-of-sample. It is not curve-fitted. It is not a fluke. The edge carries forward into unseen data.

This is what separates a real filter from a mirage – and it is the same principle behind proper robustness testing.

Why Simple Beats Complex in Trading

I used AI to generate the conditions, but the best condition AI produced was the simplest one it came up with. There is a lesson in that.

Overfitting by design. Complex indicators with many parameters fit historical data beautifully. That is exactly the opposite of what you want. A filter that perfectly matches the past almost certainly breaks in the future.

Complexity adds noise, not edge. Markets are driven by relatively simple forces – supply and demand, fear and greed, expansion and contraction. You do not need a neural network to measure whether a bar is big or small relative to recent bars. High minus low does the job.

Regime sensitivity. AI-generated oscillator conditions were highly tuned to the specific market regime they happened to perform well in. When the regime changed, they collapsed. Bar range and ATR measure a fundamental property – volatility expansion – that persists across all market regimes.

If you are currently chasing the latest AI indicator, consider this: the best filter I found across 3,500 strategies and 100 conditions is something your grandparents could have used. It works because it measures reality, not a mathematical abstraction layered on top of reality.

How to Apply This to Your Trading

If you are building breakout trading strategies, here is what I recommend based on this experiment:

• Add a bar range or ATR filter. If a bar is unusually large or small relative to recent bars, that tells you something real about whether a breakout is likely to follow through. This one filter can dramatically reduce bad trades.
• Test across many strategies, not just one. If a filter only helps one specific strategy, it is probably curve-fitted to that setup. Use an uplift approach across a large sample.
• Stop stacking oscillators. The data is clear – oscillator-based filters actively degraded breakout strategy performance. More is not better.
• Prioritize robustness over peak performance. A filter that gives you a small, consistent edge is worth far more than one that gives spectacular results in-sample and disasters out-of-sample.

You do not need 100 indicators. You do not need AI. You need one well-tested filter that measures something real about market behavior – and then you need to trust the data.

Key Takeaways

• AI generated 100 filter conditions across six categories – only volume and moving averages showed positive uplift both in-sample and out-of-sample
• Oscillators, price action, trend, and volatility filters actually made breakout strategies worse
• The single best filter was bar range / ATR – a decades-old indicator measuring high minus low relative to recent history
• Bar range reduced max drawdown by 40% and improved net profit by 60% across thousands of strategies
• It also had the highest robustness index – meaning the edge carries forward into out-of-sample data

The irony is rich. I used AI to find the best AI trading indicator, and AI told me to use something that has been around since before AI existed. Listen to the data.

Frequently Asked Questions

What are the best indicators for breakout trading?

Based on testing 100 indicators across 3,500 NASDAQ breakout strategies, bar range relative to ATR ranked first overall. It reduced maximum drawdown by 40% and improved net profit by 60%. Volume-based indicators like the Money Flow Index ranked second. Oscillators like RSI and stochastics ranked last and actually degraded performance.

Do AI-generated trading indicators work better than traditional ones?

No. In this study, 100 indicator conditions were generated using AI (Google Gemini), then tested systematically. The winning indicators turned out to be simple, well-established measures like bar range/ATR and the Money Flow Index. The AI-generated oscillator variants performed worst. The value of AI was in the systematic testing process, not in producing better indicators.

How many strategies should you test an indicator on before trusting it?

At minimum, several hundred. Testing one indicator on one strategy proves nothing – the improvement could be random. In this study, each indicator was tested across 3,500 strategies to ensure results reflected genuine statistical edge. The Uplift Index methodology measures consistent improvement across the full sample, not just cherry-picked examples.

Why do oscillators hurt breakout trading performance?

Oscillators like RSI measure overbought/oversold conditions – they signal when a move has gone too far. But breakout trading profits from exactly those extended moves. Adding an oscillator filter to a breakout strategy effectively tells the system to avoid the strongest moves, which are precisely where the edge lives.

---

Want to test indicators and build strategies without coding? BreakoutOS is a cloud-based platform that lets you prototype breakout strategies visually, validate with proprietary tools, and deploy with confidence. See the full toolkit at breakoutos.com