Testing is unavoidable in trading.
Using an untested strategy is gambling in itself.
Testing is essential for confirming whether your strategy or system has a statistical edge, and for training consistency.
Today, I will explain how to conduct testing for those who do not yet even have a system with an edge.
What I will introduce here is only my own method.
There are many ways to test, so I hope you can use this as one possible idea.
■ Build a hypothesis from what you learn, then turn it into actionable criteria
First, the foundation of a strategy is your understanding of the market.
Your strategy needs to be connected to the principles of the market.
As you study the market, you take the questions and hypotheses that arise each time, then use backtesting software to verify them.
You take what you learned, turn it into a hypothesis, break it down into actionable conditions, and then check what happens when you apply it in practice.
■ Example
For example, suppose you learned Granville’s rules and want to check whether a strategy that produces profit over time can really be built from them.
You narrow the test down only to the pullback buying point in Granville’s rules, and then verify it using backtesting software.
This is only an example.
I am not saying that this strategy has an edge.
If you are studying trendlines, replace this example with trendlines in your own mind.
First, you build a hypothesis.
For example, you might form a hypothesis like this.
“When the 20-period moving average is pointing upward, it means that, over the recent price action, the balance between sellers and buyers is tilted toward buyers, and the moving average is rising precisely because buying is actually outweighing selling.
If that is the case, other traders who missed the move may be thinking, ‘I want to buy if it gets cheaper,’ and when price gets close to the moving average, they may place orders in the same way.”
You then test that hypothesis.
In other words, anything that contradicts this hypothesis is naturally removed from the conditions.
For example, if the angle of the moving average is flat or pointing downward, it no longer fits the hypothesis of trying to collect orders from other traders who want to buy a pullback in a strong upward move.
So those cases are removed from the conditions.
Again, I am not presenting this as my own opinion, and I am not saying this has an edge.
Please do not misunderstand this point.
If the hypothesis changes, the conditions will change accordingly.
Once the hypothesis is set, you decide the conditions that need to be defined in order to fill in the required items for the experiment.
If you are trying to buy a pullback when price approaches an upward-sloping 20-period moving average, you then think about what else needs to be defined.
You cannot test with only an entry point.
You must clearly define every necessary condition.
At that point, having a hypothesis makes it harder for contradictory conditions to coexist.
Many people do not build strategies “based on a hypothesis.”
As a result, contradictory conditions sit side by side, or they try to adjust things by randomly changing parameters, and no matter how long they continue, it never works.
That is like being blindfolded and trying to pick out one single correct grain of sand from every grain of sand on Earth by pure guesswork.
There is no point in mimicking only the surface of the way someone on YouTube reads charts, then making all kinds of changes just because it is not working well.
If you do not correctly share the hypothesis that the person who built that strategy was using, you will have a hard time noticing contradictions, and it will be impossible to truly sharpen that strategy.
That is why the hypothesis is extremely important.
Also, this is surprisingly easy to overlook, but it is important to pay attention to the time displayed on the chart in the backtesting software, narrow your test down to the time periods when you can actually trade in real life, and imitate conditions that are similar to the trading you will actually do.
Testing trades that you could not actually take in real life is meaningless.
A test with no reproducibility is meaningless.
Many people only test entry rules and exit rules that they decided casually, so the test ends up having no reproducibility.
This is not about turning every detail of the entry and exit rules into numerical values.
Detail does not mean numerical precision.
It means whether the necessary conditions are placed at the necessary points, at the necessary timing.
Defining the necessary conditions in the necessary places is extremely important in testing.
And here again, having a hypothesis first is extremely important.
These details can become very extensive, so if you want to learn exactly what kinds of conditions should be defined, please see my content, TRADING SYSTEM ARCHITECTURE.
■ The first phase of experimentation
As I said earlier, there are many necessary conditions beyond just the entry and exit conditions.
You convert all of them into clear criteria, then repeat that sequence of actions.
First, repeat it about 100 times and check whether profit remains.
And this 100-trade test is only a provisional test.
It is not the full test.
With a very small sample size of around 100 trades, you cannot know the true performance of a strategy.
This is the phase for making a provisional judgment about whether it is worth moving on to full verification.
It is important to do these tests manually, not automate them.
You are the one who will actually trade, so you need to train your eyes to recognize rule-based situations from the chart.
It is important to test “the result of what you judged with your own eyes and repeated with your own hands.”
■ The second phase of experimentation
If you repeat it 100 times and profit remains, then based on the balance of risk-reward, win rate, and other factors, the strategy may have an edge.
From there, you test it more than 1000 times.
Again, a small sample size of around 100 trades is extremely small from a probability standpoint.
So you do not know whether there is an edge.
All you can say is that “there is a possibility.”
It is dangerous to conclude that there is an edge from only around 100 trades.
If you think, “I won 60 times out of 100, so my win rate is 60%,” you will inevitably be disappointed.
With such a small sample size, losing streaks and drawdowns that you have not yet encountered still exist.
Even when designing position size, how large a sample size you tested across becomes extremely important.
So you should treat a small sample test of around 100 trades as the first filter, and then test more than 1000 times.
For how to handle the insights, improvements, and refinements you gain through this series of tests, please see TRADING SYSTEM ARCHITECTURE.
■ Trial and error
You continue this cycle over and over while going through trial and error.
For example, ideas like the following may appear.
Suppose you turn pullback buying into a rule and repeat the experiment, then discover a pattern where the win rate drops especially at pullback buying points after a large extension.
After a large extension, traders who are holding profits from the prior rise are more likely to place a large amount of sell orders to take profit, so price may become less likely to rise again.
If you can form that kind of hypothesis, the next thing to verify is what happens if you limit pullback buying only to the third wave of Elliott Wave.
Or suppose you are a 1-hour chart trader, and you thought it was a Granville pullback buy against the moving average, but when you check the 4-hour chart, which is the higher timeframe, you realize that it is actually against the 4-hour trend and is a pullback selling point on the 4-hour chart.
The 1-hour traders want to buy, but the 4-hour traders want to sell.
If you understand that trader intentions are crossing there and price action is likely to become unstable, you then check what happens if you remove those trade conditions.
You might also test what happens if you limit entries to pullback buys after price breaks a higher timeframe trendline, or how the win rate changes when you change the assumed risk-reward ratio.
The deeper your understanding of the market becomes, the more ideas you will likely generate.
You test each of them one by one.
If you change even one thing, you start again from the initial 100-trade test.
Removing the cause of those losses does not mean only the profits will remain.
By removing that cause, you may also remove entries that you originally would have taken, or you may affect the win rate and risk-reward.
That changes the total result.
Even if the change is small, whenever you make any change, you must always restart the test from the beginning.
In this way, trial and error takes a very long time.
In a probability game, a large sample size is important, so experimenting with a strategy inevitably takes time.
But this is something you must spend time on.
■ Set your position size
Position size is not something you set based on how much you want to earn.
A large sample size is necessary for probability to function, but it is meaningless if you go bankrupt while collecting that sample.
In order to run your system safely over the long term, you need to calculate it mathematically “based on the performance of your strategy.”
Because you have tested across a large sample size in advance, you can calculate based on performance with higher accuracy.
The 2% rule that someone mentioned in a book has nothing to do with your strategy.
It was not mathematically calculated from your strategy either.
That is why it is important to calculate from your own tested performance.
For the exact position sizing formula I use, please see TRADING SYSTEM ARCHITECTURE.
■ Summary
First, build a hypothesis.
Define that hypothesis as repeatable conditions.
Phase one testing.
Phase two testing.
Position sizing.
That is the flow.
After this, there is an extremely important “practice phase,” but since the theme this time is the procedure for testing, I will leave that out.
Building a reproducible system, a system with an edge, is unavoidable in preparation.
No matter how powerful your consistency is, if your system has no edge, you will simply lose money.
And paradoxically, it is this preparation that gives you true consistency.
Trading is mostly preparation.
Your job as a trader is to take preparation seriously.
Real trading is only the final few percent.
Thank you for reading until the end.