For those of you who don’t already know, my formal background is in nutrition and exercise. Many of the papers I’ve published are meta-analyses. This is where you take a group of studies and analyze them together to look for trends.

When you do a meta-analysis, you’re putting together a statistical model. In a way, you’re trying to predict the effect of a particular variable on another variable using historical data (in this case, data from randomized controlled trials). For example, perhaps you’re trying to predict the impact of set volume changes on muscle size. You take a group of studies that have examined this and develop a statistical model that quantifies how a change in set volume relates to changes in muscle size.

What’s this got to do with trading?

When you are trading, you are making a prediction. You are using a number of variables such as price action and volume to predict what you think a stock is going to do. You might also use technical indicators such as RSI or stochastics. Regardless of what you use, the fact is you’re making a prediction. Whether you know it or not, you’ve developed a model for trying to predict what a stock will do. That doesn’t mean your model is a good model, but it’s a model nonetheless.

You’ve developed a model for trying to predict what a stock will do.

Now, in my scientific papers, I’ve been able to quantify how good my models are. I can give you actual numbers and the levels of uncertainty associated with those numbers. For example, I found that if you ate more than 0.5 grams of protein for every pound of your body weight, on average you would retain 2.7 more pounds of lean mass over a 12+ week period than if you ate less protein. The level of uncertainty, provided by the confidence interval, is about 1.5 pounds in either direction. Since this range (1.2 pounds to 4.2 pounds) is above zero, it tells me that protein intake is a good predictor of lean mass retention during weight loss. Also, since this range isn’t huge, it tells me my prediction is reasonably reliable. If the interval was from 0.5 pounds to 8 pounds, then my prediction is obviously much less certain.

When it comes to trading, are you able to quantify how good your model is? What may seem like a good trading model may not be so good if you start to analyze it. I’ve done this with some trading ideas and I found that they weren’t nearly as good as I had originally thought.

…are you able to quantify how good your model is?

In another article I talked about back-testing and real-time testing your trading system. When you do this, the question to ask yourself is this: is this system reasonably reliable for predicting how a stock will behave? How does it work in different market conditions and over time? Does it make logical sense that this system would work? Is there a rationale behind why it should work?

The results may surprise you if you start to take a close look at these things. I’ve found that most technical indicators that traders use are very poor predictors of how a stock will behave. Things like RSI or MACD are lagging indicators and simply reflect what a stock has done in the past. However, when you analyze them closely, you find that they have little to no predictive value in most instances. There is no logical or rational reason as to why these indicators would help predict how a stock will behave (which is what you’re truly interested in).

…most technical indicators that traders use are very poor predictors of how a stock will behave.

A perfect example is Fibonacci retracement levels. These are pretty much worthless for predicting stock movement. When you think about it, there is no logical or rational reason why a stock should retrace to these predefined levels. When you analyze these retracements more closely, you’ll find that there are numerous instances where a stock doesn’t behave according to these retracements. In fact, there are enough instances that it gives this indicator questionable value in making predictions.

Another example are the pivot points and associated resistance and support levels R1, R2, S1, and S2. Is there any logical or rational reason why a stock should behave according to these levels? Of course not. I’ve studied numerous stock charts and found that these levels often don’t mean anything and have little predictive value. They simply provide an arbitrary framework upon which traders make decisions.

When you have a model that has some sort of logic or rationale behind it, then the reliability of your model increases. For example, earlier I had mentioned a statistical model in the field of exercise where more set volume predicts greater increases in muscle size. There is a physiological rationale behind why this would be the case.

This is true with trading as well. For example, one of my strategies is shorting stocks that have run up way too far, way too fast. There is a logical reason behind why this would work that has to do with a combination of human psychology and basic supply and demand. Some traders like to trade around round $ marks and there is logic behind this too. Humans like round numbers and they always will. Since humans make up the markets, these numbers will often result in approximate support and resistance points for this reason.

I’m a scientist at heart, and I approach my trading like one.

I’m a scientist at heart, and I approach my trading like one. I test my ideas to see their predictive value, similar to testing a hypothesis. I’ve thrown away most technical indicators because they offer little predictive power over time. I ask myself, “Does this trading idea make logical sense? Is there an underlying reason why this would work? Are the odds consistently in my favor when I trade this method? What is the level of uncertainty associated with this method? Is this level of uncertainty acceptable?”

Be sure to ask yourself these questions as you develop your own trading system.