Research Article #35 - Building Robust Trading Systems: The Importance of Signal Validation
Testing and Verifying Our Trading Signals Effectively
👋 Hey there, Pedma here! Welcome to the 🔒 exclusive subscriber edition 🔒 of Trading Research Hub’s Newsletter. Each week, I release a new research article with a trading strategy, its code, and much more.
Join over 4K+ readers of the newsletter!
I’ve been reading Robert Carver’s book “Systematic Trading: A unique new method for designing trading and investing systems.”
By the way, I highly recommend it for anyone that’s serious about systematic trading.
Have you ever wondered if your signal makes sense?
We all seen a trading rule online, that sounded interesting (or not), like buying an asset when the RSI is above 70, and we wonder if that really works.
No, plotting the returns is not a good method of validation.
I can generate random data, get thousands of different classic technical signals, and find a profitable trading strategy, on that completely random data.
Finding historical profitable strategies is the easy part.
With the technology we have today, we can generate hundreds of those “strategies” in a few minutes.
This is where the research process gets fun.
How do you validate your trading rules, to ensure that your system has the highest odds to succeed once deployed?
That’s we will get into today.
So let’s get to this!
Index
Introduction
Index
Picking Trading Signals
Removing Signals
Testing Statistical Significance with T-Test
Portfolio Allocation
Strategy Analysis
Python Code
Picking Trading Signals
With the access we have today to AI, machine learning and all the technology being built, it’s easy to try and be smarter than we should.
There’s nothing wrong in trying things out, but in general, keeping things simple pays dividends in markets.
Why is that? There’s a few reasons:
Financial data is very noisy and it’s really hard to find real signal.
Limited amounts of data make it hard to have statistical significance.
Everyone’s looking at the same data, if there’s something obvious, it’s already been found.
The most common pitfall is trying to find the best possible variation of a rule.
Take the example of someone that wants to trade an SMA crossover. What variation should he use?
10/20 crossover?
20/40 crossover?
40/80 crossover?
Maybe we can run a machine learning model and find that actually the 47/65 crossover is best.
There’s nothing inherently valuable about that signal, it just happened to work best in the data you have, most likely out of pure chance.
So here’s my process to pick signals: