How to Backtest a Trading Strategy Like a Fool

I woke up this morning and grabbed a cup of coffee, then just for fun, I decided to type "how to back-test" into YouTube.

I was curious to see how much more advanced the younger generation was at systematic trading.

The results for me were like fingernails on a chalkboard...

Only trade what you test - or you are a fool!

Backtesting is, of course, how you verify your trading rules will make money.

If your rules don't make money in a simulation, they aren't going to magically start making money in the real world.

A properly tested trading system has data you hold back ("out-of-sample") from the simulation.  

Next, you run your rules on the out-of-sample data and hope it looks like the in-sample data:

All the top videos were of guys and gals dragging their charts across the screen and "backtesting" by hand.

I wanted to reach through the computer and slap them silly.

That's not how to backtest!

It's the blind leading the blind out there.

People literally believe playing connect the dots with trend lines is how you make money in the markets.

I explained this to my buddy, and he believes that if enough people say it on the news, then it must be true.

No wonder there was so little progress until the scientific method.

That's where people started thinking in terms of evidence rather than consensus.

Einstein's Theory of Relativity was crapped on by 100 scientists.

They sent him a letter telling him how wrong he was.

To add extra clout to the message, they all signed it.

Albert's response: "Why 100? If I were wrong, one would have been enough."

Meaning that all it takes is one piece of evidence to debunk his theory.

If I received a document signed by 100,000 people saying the earth is flat, I'll find their leader and park his fat ass next to Bezos in the next launch of his phallic-shaped rocket.

It's like American Airlines finding the shortest people, and asking them what they thought about the legroom.

OK, ok, I'll climb down off my soapbox for a minute.

Let's talk about how to backtest today and how it helps mint millionaires and a few billionaires.

First of all, it's just plain lazy not to have a computer do the backtesting for you.

According to Wikipedia, humans have a TON of biases.

One of them is ignoring losing trades when you wing it and backtest by eye.

I'm guilty of this kind of backtesting decades ago, and it cost me not only the money I lost but the money I could have gained from implementing a proven strategy.

I didn't just lose $20k, I probably lost millions in opportunity over those years.

I wandered aimlessly from technique to technique.

When you set up a few trading rules, the next step is to find the optimum settings for trading indicators.

What most fail to do is hide a portion of the historical data from the computer.

That's to confirm you have a real edge in whatever you're trading, and not over-fitting to the past.

It's insane that some platforms with millions of users don't have any out-of-sample testing capability.

Any strategy built without out-of-sample testing is immediately suspect.

Another tip on how to backtest even better is to have multiple areas hidden from the optimization process.

If you have a big hunk of recent historical data that's unseen by the computer, you'll probably have the wrong settings to maximize profits.

That's because markets drift over time.

But by breaking your test into several different sections, you'll likely boost performance considerably.

That's like adding high-octane fuel.

Here's that being tested on my Slugger trading system:

Here's a hot tip if you've made it down this far.

Traders are leaving gobs of profits on the table by not including "portfolio ranking" in their trading systems.

That's where you have a basket of stocks...like the entire NASDAQ 100...and you just trade the top 10.

You could rank by highest momentum over the past few months -- that's a good starting place.

Like the Billboard Hot 100 ranks by highest sales.

Most trading platforms have at least a million people are using them.

This makes it damn near impossible to discover edges that aren't based on the price of a single symbol.