Beyond Binary: Rethinking Trading Indicators Through The Lens Of AI Research
When Binary Thinking Meets Fluid Markets
Important: There is no guarantee that ATS strategies will have the same performance in the future. I use backtests and forward tests to compare historical strategy performance. Backtests are based on historical data, not real-time data so the results shared are hypothetical, not real. Forward tests are based on live data, however, they use a simulated account. Any success I have with live trading is untypical. Trading futures is extremely risky. You should only use risk capital to fund live futures accounts and if you do trade live, be prepared to lose your entire account. There are no guarantees that any performance you see here will continue in the future. This is why the best way to trade is with a simulated account on live data. I recommend using ATS strategies in simulated trading until you/we find the holy grail of trade strategy.
As a quick reminder, we’re on the hunt for the holy grail of automated trading strategy. If you have any questions, start with the FAQs and if you still have questions, feel free to reach out to me directly at AutomatedTradingStrategies@protonmail.com.
Let’s get into it…
As you all know, I’ve been obsessed with Large Language Models (LLMs) lately. Recent advancements have demonstrated remarkable capabilities in contextual analysis, suggesting potential applications in market behavior interpretation. This is why I’ve been spending a lot of time trying to understand how LLMs “think”. Understanding these models serves two critical purposes: 1) to optimize prompt engineering for maximum effectiveness, and 2) to enhance the contextual awareness of ATS systems.
In my research, I came across the paper: "Are Emergent Abilities of Large Language Models a Mirage?" Also given the "Best Paper Award" at the 2023 NeurIPS (Neural Information Processing Systems) Conference, the paper argues that "emergent abilities" in large language models (LLMs) might be an illusion caused by the way researchers...(wait for it)...measure performance.
I wasn't expecting that.
It finds that what appears to be sudden "emergent" behavior may actually be smooth, continuous improvements. It goes on to say that using different metrics on the same model outputs can make emergent abilities either appear or disappear. In other words, emergent abilities might be a mirage caused by using the wrong benchmark metrics rather than fundamental changes in the model.
It’s that last part that caught my attention because it has huge implications for our hunt. Just as traders watch for false breakouts in price action, these researchers spotted something similar in how we evaluate LLMs — they found that what looked like breakout performances might actually be false signals created by our measurement tools. By extension, the indicators we use might be masking smooth market movements by using threshold-based or discontinuous measurements.
The paper demonstrates how switching from discontinuous metrics (like pass/fail tests) to continuous ones reveals smooth performance improvements. This suggests that those strategies that use binary trading signals (like crossing above/below a threshold) might be artificially creating "sudden" signals when the underlying market movement is actually gradual, which could be leading to two types of errors:
False Signals: These are the head-fakes of the trading world—when indicators trigger an entry or exit, but the expected move never materializes. Think of those times when RSI shows oversold, but the up/downtrend continues r-e-l-e-n-t-l-e-s-s-l-y.
Missed Moves: These are the trades that got away when indicators fail to trigger on legitimate opportunities. Like waiting for that "perfect" moving average crossover.
Both scenarios hit our trading accounts in different ways: one through direct losses, the other through opportunity cost. This insight from LLM research has huge implications for our hunt. While we've been focusing on finding the perfect combination of indicators, perhaps we should instead be questioning how we're measuring market movements in the first place.
What Are The Best Indicators To Use
This isn’t the first time this has been studied. Anyone in data science is most likely aware of more seminal studies like Neftci's work on continuous indicators (1991). It revealed something every trader needs to understand: binary trading rules (think simple crossovers) often miss the market's underlying structure— like trying to track a complex trend with just support and resistance lines. His research showed that probabilistic indicators, which read the market more like a moving average than a simple price cross, delivered more reliable signals.
Fast forward to 2000, and we see Lo, Mamaysky, and Wang's landmark study, "Foundations of Technical Analysis," confirming this thesis. They discovered that many of our favorite chart patterns might be more like mirages in the desert—artifacts of how we're measuring rather than true market signals. Their work demonstrated that continuous pattern recognition produces more actionable trading insights.
Through this research, two major types of problematic indicators emerged:
Nonlinear metrics: These are like using multiple time frame analysis where the relationships between signals aren't proportional
Discontinuous metrics: These create binary yes/no signals, similar to how a simple price cross of a moving average either triggers or doesn't, with no middle ground
Just as a choppy market can create false breakouts, these indicators can create an illusion of sudden pattern changes when the underlying market movement is actually smooth and continuous.
So what? The next section examines how these insights apply to ATS strategies.
What Does This Mean for Your Strategy
Let's chart out the practical implications for automated trading systems. Just as experienced traders know that noise in the market can create false signals, what these papers are telling us is that certain types of indicators might be adding unnecessary noise to our analysis. When your RSI and Double Stochastic are showing overbought, but your ADX is still trending strong — which signal do you trust?
This dilemma explains why smoothing indicators like moving averages are so popular in trading systems. However, every trader faces the classic smoothing trade-off: less noise means more lag, and more lag means missed entries and exits. It's like trying to catch a breakout—wait too long for confirmation, and you've missed the damn thing.
Here's where the research gets interesting. By carefully selecting and combining specific indicators in a particular way, we might be able to achieve what every trader dreams of: the noise reduction of smoothing without the delayed signal.
Imagine catching trends early while still avoiding false breakouts. This could be a game-changer for how we:
Identify valid trade setups earlier
Filter out false signals more effectively
Optimize entry and exit timing
Now I want to take this a step further by discussing the practical implications of this research and the specific indicator combinations you can use to improve your trading. I think you’ll be surprised at how simple some of these changes are, but they are nuanced. We are about to get very detailed because in a non-binary world, the sequence that you do something and the degree with which you do it matter.
Before we dive into specific examples of problematic indicators and their alternatives, I want to share something that helped crystallize these concepts for me. I've curated a podcast reviewing the "Are Emergent Abilities of Large Language Models a Mirage" paper, specifically tailored for traders. The insights from Gemini about pattern recognition are particularly insightful.