Transcription
I tested 100 trading indicators, and I found this.
Now, in my earlier years of trading, I've wondered, and maybe you've wondered as well, what actually makes an indicator profitable on the live markets. Now, I used to assume it was all about finding the right indicator, the one secret signal that would unlock profitability. But after testing and using hundreds of indicators, consuming countless videos, books, and even academic papers, and also trading for the past 6 years, I've made some surprising conclusions, like how the indicators everyone thinks won't actually work actually do. And how the real edge comes from something most traders have never heard of in the first place.
So, what does actually testing hundreds of, you know, trading indicators look like? Well, it doesn't look like this. I don't just put them all on one chart. Each one I have either run live or I've tested historically using a backtesting process. This includes fully written, uh, up via code. So no human bias, no manual backtesting, none of that kind of stuff. I've tested it on many years of historical data. And I've been using robustness techniques to make sure the results weren't just noise.
Now, if you want to learn more about backtesting, feel free to check the first link in the description for a free cheat sheet. So, let me show you what I've discovered after testing hundreds of indicators and why the answer isn't what you think.
Now, the first one is simple indicators just work better. Simple indicators have consistently outperformed complex ones. At least from my testing. I'm talking about moving averages, RSI, Bollinger bands, Keltner channels, for example. All the basic stuff that everyone says, well, everyone knows, so there's no way it could work. Versus things like machine learning models using neural networks, custom-coded multivariable indicators using 15 parameters. Over the past 16 years, I've tested everything from LSTMs predicting price action to very simple indicators like moving averages. And nine out of 10 times the simple indicator won.
Why? Because you're going to run into this, which is overfitting. Now, I've explained overfitting in many videos, many, many times, but as a very short recap, primarily this is where your indicator or your strategy or your backtesting becomes too tailored to the historical data. So it won't actually work live on new, you know, out-of-sample or general data. As a visual guide, you can kind of see this, right? So we have these dots on this graph. The underfitted version is just going straight through them, right, and it doesn't really line up with the dots. The good fit or robust fit is generally in between them, but it's not touching every single dot and it's not the perfect fit. Whereas overfit, it here, as you can see, it's touching every single dot primarily, and and that's kind of following that path. Imagine that with your historical data. An easy example is let's say I did, you know, a backtest and the moving average 22.59 worked the best. It has amazing results. That most likely is historically overfit because there's no reason why the 22.59 is going to work the best. And especially if I change that slightly and then performance drops, then I'm pretty confident that it's overfit.
Now, while simple indicators can help you avoid this trap, obviously you can always overfit anything as long as you, you know, torture it enough to give you the answer you want, but most of the time when I'm using a simple indicator, I'm I'm just running one walk-forward optimization or simple backtest and I can get the general idea if it works or if it doesn't work.
Now, this leads me into the next kind of main discovery that I think is important to understand, and that's the four main types of indicators. Now, here's what most traders miss is that it's not just about using simple indicators. It's about understanding what type of the indicator is and when to use it and also how it really works.
Now, there are four main types. You've got trend indicators. So this is like moving averages, ADX, etc. This normally just tells you the direction and strength of the trend. Then you have momentum indicators. So like RSI and Momentum. This tells you the speed of price movement and also kind of the strength of the trend as well. These two are normally kind of combined together, but you can sort of separate them. Next, you've got volatility indicators. So things like Bollinger bands and ATR. ATR is Average True Range. These are best for breakout strategies or position sizing. I use ATR for pretty much all of my strategies for position sizing. This tells you whether there's an expansion or a contraction of price range. Next, you have volume indicators. So, you have OBV and like Volume Profile, etc. These are best for normally confirming price moves and filtering, and they also tell you sometimes the strength behind the move.
Now, the key here is matching indicator types to the strategy type and understanding them fully. So, for example, if I'm building a trend-following strategy, I want to be using trend and momentum indicators most of the time. So, that's moving averages, etc. Now, these can overlap. So, for example, Bollinger bands, which is kind of a volatility indicator, you can use it as a trend indicator in some essence. For example, when price breaks above that Bollinger band, that is now, you know, a trend signal to go long, for example, because it's breaking out. And these apply to the other ones as well that can sometimes be overlap, but it's good to really understand what it's measuring and how it's doing.
So, now, next is the best method. And this is the main thing I think I want people to learn from this video. We're going to go over something called an ensemble method.
Now, an ensemble method was primarily used in machine learning, and it was primarily, um, used kind of like in this case, right, where you have multiple different models, and they go into like a meta-learner or the main model, and then that gives you a final prediction. Now, we're following a similar structure here, but obviously, we're not applying machine learning to this. You can obviously, if you wanted to, but here's the main concept. I know it can sound super complicated, but the idea is actually quite simple. Instead of relying just on one indicator to make a decision, you want to combine multiple indicators into one forecast.
So, this is kind of a rudimentary image here. Uh, but primarily you can see, right, there's three different boxes, and you can see one box is is red, then the next one is green, etc. Now, the highlighted section here is when all boxes are green, meaning when all of them are showing, uh, that you want to go long, that it's then going long. But for example, if there was a mixture, it was balanced, or it wasn't fully even, then you wouldn't be going long or you wouldn't be going short. And for example, over here, where they're all showing red, you'd be going short. This primarily just helps you combine multiple indicators into one forecast so you can get a good idea of the direction and strength.
Let me give you a practical example of how this works. So, let's say we have each indicator, and we're going to give it a score from minus one to one. Minus one here equals a short signal, zero equals neutral, and plus one equals a long signal. Then you add them all up. So, let's say I'm using five indicators. I'm using RSI, MACD, ROC, uh, Stochastic, and a moving average slope. I have one for RSI, one for MACD, one for ROC, zero for Stochastic, and then moving average slope. I have one as well. So, in total, I have four because I'm adding up all these ones or I'm just adding up everything primarily. And I'm dividing that by five because there's five indicators. And I get a score of 0.8. Now, this would be considered a somewhat strong long signal, right? Because it's positive and it's also near one, which would be the 100%, let's say.
Now, this is so powerful because let's say you have a single indicator and you're just saying when RSI says long, you're going long. But RSI signal strength might be wrong 30% of the time, right? And you're going to be taking full losses on those wrong signals. Whereas with an ensemble method, you can get a better kind of vary of the land. You can avoid bad trades because the ensemble has a low confidence if other things are conflicting.
Now, obviously, you do want to be using indicators that are measuring either similar things or, uh, a similar strategy type, for example, like trend following or reversion, etc. You don't want to really be combining those because then it's always going to be somewhat mixed, right?
Now, I 100% didn't come up with this method. This method was invented way before, at least me focusing on it. And actually, there is a great book called "Advances in Financial Machine Learning" by Marcos Lopez de Prado, I think, and he dedicated a whole entire chapter to this ensemble method. He states that an ensemble method combines a set of weak learners to create a strong learner that performs better than any of the individual ones. Ensemble methods help reduce bias and variance. The results often lead to much smoother equity curves for performance, lower drawdowns, and better risk-adjusted returns, and most importantly, more robust to changing market conditions.
Now, why would he say weak learners? Well, what a weak learner is, is kind of what I'm mentioning here with the simple indicators. Most of the time, these overly complex indicators, maybe they do perform, you know, sometimes better depending on how smart you are and how you created it. But the easier method is really just to get a lot of these very simple indicators that have worked over time, but they've degraded in performance. So, like moving average crossovers, Bollinger band breakouts, etc., you could put them all into kind of one soup, combine that soup, and now you've got like a way better, uh, soup, let's say, and it's a way better tasting, way better flavor, etc. It's just a better performer because you're combining all of them into one.
This is actually something, you know, I do myself in some of my own strategies, and it can get really complex if you want to go there. But overall, the idea is quite simple.
Also, this leads to one extra thing, and it's probably another huge addition to these sort of strategies, is that you can then become continuous. What continuous means is you can size based on the strength of that signal. So, it no longer just has to be that you're looking for, you know, uh, higher than, let's say, 0.5 or whatever your threshold is to go long or to go short. You could then size based on what number you got. So, let's say you got, you know, 0.8, uh, as we did in this example, you could then size 80% of your whatever amount of risk you want to put on for that particular trade. Whereas, if you got 0.2, it would be 20% or whatever you want to do with that.
Now, there's multiple ways to, you know, do position sizing with that, but that ultimately leads to the most efficient strategies, which are normally the higher Sharpe ratios and normally have the smoothest looking equity curves.
If you want to learn more about this, feel free to check out the backtesting boot camp. But I hope you enjoyed this.