How to Build a Trading System That Actually Works
Most traders skip these five steps. Here's the right way to do it.
So you want to build a trading system. Maybe you’ve already tried. You found an indicator that looked great on NVDA or TSLA, ran a backtest, got excited about the results, and then watched it fall apart the moment you traded it live or tried it on anything else.
That’s not bad luck. That’s a process problem.
Whether you’re trying to build a mechanical trading system for the first time or you’ve been down this road before, the mistakes are almost always the same. People skip the boring foundational stuff and jump straight to the fun part, which is running backtests and watching equity curves go up. It feels like progress. But you’re basically building a house without a foundation.
Algorithmic trading and system building look complicated from the outside, but the core process is actually pretty straightforward. There are five steps, and most people only follow two or three of them. Here’s what the full process actually looks like.
Step 1: Sit With the Data Before You Do Anything Else
Most people come into this with an idea already in their head. They read something online, heard about a setup that “always works,” and now they want to test it. So they go hunting through the data for proof that they’re right.
That’s backwards.
Before you form any strong opinions, just look at the data. Pull up charts across different years, different market conditions, calm periods and crazy ones. Watch how prices move, where volume spikes, what happens after big drops. Don’t try to find anything yet. Just look.
One tool I use for this is the Genetic Evolver in Wealth-Lab. It’s essentially a way to let the software loose on a large dataset and surface small edges you might never think to look for on your own. It’s not about handing the wheel over to an algorithm. It’s more like using it as a research assistant to point you toward areas worth digging into. You still have to make sense of what it finds.
This is the part nobody wants to do because it feels like you’re not making progress. But what you’re really doing is building intuition. You’re learning what the market actually does, not what you think it does. That’s the whole game.
Step 2: Come Up With a Simple Hypothesis
After spending real time with the data, something usually starts to jump out. A pattern that repeats. A behavior that shows up consistently. A tendency.
That’s your hypothesis.
It doesn’t need to be complicated. In fact, the simpler the better. Something like: “When the market sells off hard on big volume, it tends to bounce within a few days.” Or: “When volatility is low and the market starts trending, it tends to keep going.”
The key word there is “tends.” You’re not looking for something that works every single time. Nothing does. You’re just looking for a repeatable edge that you can actually explain in plain English.
Here’s a useful test: try explaining your hypothesis to someone who doesn’t trade. If you need a chart open to get through it, the idea is probably either too complicated or not fully formed yet. That matters, because if you can’t explain why something works, you won’t know when it stops working.
Step 3: Write Down the Rules Before You Touch Your Software
This is where things get real, and where a lot of people get sloppy.
Take your hypothesis and turn it into an actual set of rules. When do you get in? When do you get out? How do you size the position? What happens if it goes against you right away?
Be specific. The goal is to write it out clearly enough that someone else could follow your rules and make the exact same decisions you would. No guessing, no interpreting.
And keep it simple. Seriously. Every extra rule you tack on is another way for the system to fit the past instead of capturing something that will actually repeat. Simple systems are almost always more durable than complex ones.
The habit worth building here: write the rules out in plain language before you ever open your backtesting platform. If you can’t describe the logic without a screen in front of you, go back and think it through some more.
Step 4: Test It to Learn, Not to Confirm
Here’s where most people go off the rails.
They run a backtest hoping to see a beautiful equity curve. If it looks good, they declare victory. If it doesn’t, they start tweaking things until it does. A rule gets adjusted here, a parameter gets tuned there, and suddenly the backtest looks great.
But now you haven’t tested a hypothesis. You’ve just reverse-engineered the past. That’s called curve fitting, and it’s one of the most common reasons systems fall apart in live markets.
The real purpose of a backtest is to understand how the system behaves, not to prove it works. You want to know where it struggles. What does a bad stretch look like? How deep do the drawdowns get? Does it hold up across different years or is all the performance stacked in one particular period?
A system that only shined during the 2020 crash or the 2013 to 2019 bull run isn’t a robust system. It’s a system that got lucky in a specific environment.
One thing that helps: write down what you expect to see before you run the test. If your hypothesis is about mean reversion after big drops and the backtest shows it only works in trending markets, something doesn’t add up. Either the rules don’t match the idea, or the idea isn’t what you thought it was.
Step 5: Honestly Evaluate What You’ve Got
You’ve got results. Now comes the most important part of the whole process, which is being honest about what you actually have.
Start by widening the scope. If you built and tested this on one symbol, take it out for a longer walk. Does the same logic hold up on other ETFs, other markets, other sectors? A system that only works on one ticker is a fragile thing. Maybe it’s capturing something real, or maybe it just fit that one ticker’s history. The only way to find out is to test it more broadly.
Then ask some harder questions. Is the performance consistent year over year, or is it basically one or two great years carrying the whole backtest? How bad do the drawdowns get, and how long do they last before recovering?
And if you already have other systems running, here’s a question worth thinking about: does this new one behave differently from what you already have? Pull the equity curves up side by side. Do they tend to hit rough patches at the same time, or do they struggle independently of each other? You’re not chasing perfect non-correlation, that’s basically impossible to achieve in practice. You’re just looking for something that adds a different flavor to what you already have. Two systems that don’t both blow up at the same time is a genuinely good thing.
If the system holds up across all of that, it earns a spot in your research. If it doesn’t, you go back to step one. No shame in that. Most ideas don’t survive this process, and that’s exactly the point.
Over time you’ll build a core set of principles that tend to be advantageous across symbols and timeframes. These can become plug and play into future data sets and watchlists.
This Is a Loop, Not a Checklist
The thing people don’t tell you when you start down this road is that you never really finish. The process doesn’t end when you find something that works.
Markets change. What worked three years ago might quietly stop working tomorrow. The traders who stick around are the ones who treat this as an ongoing practice, not a one-time project. You’re always somewhere in the loop.
With the systems I provide publicly, that’s the way I think about it. Every system goes through this process. No shortcuts, no skipping the boring parts.
If you’re just getting started, that’s actually the best possible position to be in. You get to build the right habits from the beginning. Start with the process, and the rest gets a lot easier from there.
Have a Great Night!
Dave Johnson