Is PF1.37 an excellent EA?
## Five Numbers I Check When Looking at Backtest Results
When evaluating an EA (automatic trading system), one of the numbers you often see is “PF (Profit Factor).”
PF1.2, PF1.5, PF2.0――。
The larger the number, the more effective the EA appears, and when I first started making EAs, I was focused on increasing PF.
However, after repeatedly backtesting and adjusting parameters over several years, I began to feel that evaluating an EA based only on PF is quite dangerous.
This time, I will introduce what parts of the backtest results I actually look at when developing and validating an EA.
---
## What PF actually is
PF (Profit Factor) is calculated as
**Total Profit ÷ Total Loss**
For example,
- Total Profit: 1,370,000 yen
- Total Loss: 1,000,000 yen
Then,
**PF = 1.37**
PF above 1.0 means that, at least in backtests, profits exceed losses.
Therefore,
you tend to think, “PF is 1.5, so it’s excellent,”
“PF 2.0 means a very strong EA.”
But there is a major pitfall here.
PF is merely the ratio of profit to loss over the entire test period and does not tell you
**how that profit was accumulated**
.
---
# The Five Numbers I Look At
When evaluating an EA, I mainly check the following five items.
1. PF
2. Number of trades
3. Maximum drawdown
4. Yearly performance
5. Profit and loss curve
I will go through them in order.
---
## 1. PF
Of course I check PF.
Personally,
**If PF can be stably maintained around 1.3, it is worth considering**
in my view.
Conversely, even if PF2.0 or PF3.0 appears, I don’t get excited at that moment.
Because, by finely optimizing parameters, you can relatively easily push PF up for a specific period.
What matters is
**not how high PF is, but how stable the PF is**
, I believe.
---
## 2. Number of Trades
Next, I look at the number of trades.
For example,
- PF1.80
- 30 trades
vs.
- PF1.37
- 300 trades
In that case, I place much weight on the latter.
If the number of trades is small,
PF can be high simply because a few lucky trades happened to be profitable.
Conversely, if PF around 1.3–1.4 is maintained over hundreds of trades, the statistical reliability is relatively high.
In one EA I’m validating, there are
**about 300 trades from 2021 to 2026**
that occurred.
With this many trades, it’s hard to claim profits were just from a few lucky trades.
---
## 3. Maximum Drawdown
Personally, I value this even more than PF—the
**Maximum Drawdown (Maximum Drawdown)**
For example,
- PF1.60
- Maximum DD 45%
If you were operating with 1,000,000 yen,
you might temporarily see capital drop to around 550,000 yen.
In backtest screens,
you tend to only see that you ultimately made a profit, but when actually operating, whether you can endure the drawdown along the way is extremely important.
I personally
**focus more on avoiding large drawdowns than increasing profits**
because with EAs, longevity matters more than a single big gain.
---
## 4. Yearly Performance
This is especially important.
For example, if an EA is backtested from 2021 to 2025 and shows
overall PF 1.40
this may look decent.
But when broken down by year, you might see
- 2021 PF 1.13
- 2022 PF 1.33
- 2023 PF 1.46
- 2024 PF 1.26
- 2025 PF 1.80
resulting in
this kind of picture.
In this case, it is not an EA that profits evenly every year.
2021 was quite challenging, while 2025 performed very well.
In other words,
you can check whether a specific year disproportionately boosts the overall PF.
When I backtest, I first look at long-term results, then always break down by year to confirm.
---
# The Same EA wins in 2022 but loses in 2024
Sometimes there are even more extreme cases with EA development.
With a certain logic,
2022:
- PF: 1.26
- Profit: positive
but
2024:
- PF: 0.87
- Profit: negative
resulted.
The logic and parameters were the same.
Yet the results can differ greatly.
This is because the market itself has changed.
Some years see trends form more easily, others have longer ranges.
Volatility also changes.
Therefore, you must look not only at how it performed over the past five years, but also
**which market conditions it is strong in and which it is weak in**
.
## 5. Profit and Loss Curve
Finally, I always look at the
**Balance / Equity Curve**
.
I may spend more time looking at this than PF.
Ideally, the curve should be
gradually, yet surely, rising to the right.
On the other hand, EAs that
- stay flat for a long time
- suddenly surge in profits
- drop capital rapidly
- concentrate profits in only certain periods
are to be cautioned.
In particular, be careful with EAs that show
“little profit for most of the time, then a few trades yield large profits.”
PF may look excellent in such cases.
---
# Not chasing the “best parameters”
In EA development, there is something I’ve been especially conscious of lately.
That is,
**not chasing the best backtest result**
.
When you optimize, you may end up with
PF1.30
PF1.40
PF1.60
PF2.00
and so on.
You naturally want to pick the parameter with the best number.
However, that parameter may have only matched past moves by coincidence.
In other words,
**curve fitting (overfitting)**
Therefore now I look for a region where the results do not crumble even if numbers change a bit.
For example,
a parameter changes from
20 → PF1.34
21 → PF1.38
22 → PF1.36
23 → PF1.35
24 → PF1.33
If so,
you can feel fairly secure.
On the other hand,
20 → PF1.05
21 → PF1.08
22 → PF2.10
23 → PF1.03
24 → PF0.98
In that case,
PF 22 may be fitting by chance alone.
I personally would not adopt the latter.
---
# Entry conditions vs. conditions not to enter
When I started making EAs,
I used to think about, “What conditions should trigger an entry?”
But now I emphasize
For example,
- ADX is low
- EMA slope is weak
- Price is far from the EMA
- Volatility is extremely low
- Specific time windows
- Friday
- Monday morning
, etc.
Adding such filters reduces the number of trades.
In exchange, you can
avoid making bad trades.
In EAs,
reducing losing trades often improves results more than increasing winning trades.
---
# Is PF1.37 a good number?
So,
**PF1.37**
is that a good number?
My answer is
**“PF alone cannot determine it.”**
.
For example, if you meet these conditions:
- Trades > 300
- Profitable over multiple years
- Maximum DD within tolerance
- No extreme yearly bias
- Profit curve relatively stable
then PF1.37 is highly favorable.
On the contrary, if you have
- 20 trades
- Only one year tested
- Maximum DD 50%
- Profits from a few big wins
then operating in real life would be very cautious.
---
# Summary
When looking at backtests for an EA, PF tends to catch the eye.
However, after validating over a long period, PF is only one of the numbers to evaluate.
What I currently check are the five: **PF, number of trades, maximum drawdown, yearly performance, and profit and loss curve**.
Moreover, I focus on a logic where the results do not crumble significantly even if some parameters change, rather than chasing the set of parameters that yields the best results.
In EA development, it is important to consider not only how much profit you can make, but also how to avoid large losses as much as possible.
In the future, I plan to write about what I learn while actually creating EAs, such as
- where to look during optimization
- how you use ADX and ATR
- the reasons why performance changes year by year
- how to think about the required margin for EAs
- handling trend vs. range markets
I will continue to share what I learn by actually creating EAs.
Note: This article is based on the author's own EA development and backtesting experience. It does not guarantee future profits or performance. Actual trading should be done at your own judgment and responsibility.
When evaluating an EA (automatic trading system), one of the numbers you often see is “PF (Profit Factor).”
PF1.2, PF1.5, PF2.0――。
The larger the number, the more effective the EA appears, and when I first started making EAs, I was focused on increasing PF.
However, after repeatedly backtesting and adjusting parameters over several years, I began to feel that evaluating an EA based only on PF is quite dangerous.
This time, I will introduce what parts of the backtest results I actually look at when developing and validating an EA.
---
## What PF actually is
PF (Profit Factor) is calculated as
**Total Profit ÷ Total Loss**
For example,
- Total Profit: 1,370,000 yen
- Total Loss: 1,000,000 yen
Then,
**PF = 1.37**
PF above 1.0 means that, at least in backtests, profits exceed losses.
Therefore,
you tend to think, “PF is 1.5, so it’s excellent,”
“PF 2.0 means a very strong EA.”
But there is a major pitfall here.
PF is merely the ratio of profit to loss over the entire test period and does not tell you
**how that profit was accumulated**
.
---
# The Five Numbers I Look At
When evaluating an EA, I mainly check the following five items.
1. PF
2. Number of trades
3. Maximum drawdown
4. Yearly performance
5. Profit and loss curve
I will go through them in order.
---
## 1. PF
Of course I check PF.
Personally,
**If PF can be stably maintained around 1.3, it is worth considering**
in my view.
Conversely, even if PF2.0 or PF3.0 appears, I don’t get excited at that moment.
Because, by finely optimizing parameters, you can relatively easily push PF up for a specific period.
What matters is
**not how high PF is, but how stable the PF is**
, I believe.
---
## 2. Number of Trades
Next, I look at the number of trades.
For example,
- PF1.80
- 30 trades
vs.
- PF1.37
- 300 trades
In that case, I place much weight on the latter.
If the number of trades is small,
PF can be high simply because a few lucky trades happened to be profitable.
Conversely, if PF around 1.3–1.4 is maintained over hundreds of trades, the statistical reliability is relatively high.
In one EA I’m validating, there are
**about 300 trades from 2021 to 2026**
that occurred.
With this many trades, it’s hard to claim profits were just from a few lucky trades.
---
## 3. Maximum Drawdown
Personally, I value this even more than PF—the
**Maximum Drawdown (Maximum Drawdown)**
For example,
- PF1.60
- Maximum DD 45%
If you were operating with 1,000,000 yen,
you might temporarily see capital drop to around 550,000 yen.
In backtest screens,
you tend to only see that you ultimately made a profit, but when actually operating, whether you can endure the drawdown along the way is extremely important.
I personally
**focus more on avoiding large drawdowns than increasing profits**
because with EAs, longevity matters more than a single big gain.
---
## 4. Yearly Performance
This is especially important.
For example, if an EA is backtested from 2021 to 2025 and shows
overall PF 1.40
this may look decent.
But when broken down by year, you might see
- 2021 PF 1.13
- 2022 PF 1.33
- 2023 PF 1.46
- 2024 PF 1.26
- 2025 PF 1.80
resulting in
this kind of picture.
In this case, it is not an EA that profits evenly every year.
2021 was quite challenging, while 2025 performed very well.
In other words,
you can check whether a specific year disproportionately boosts the overall PF.
When I backtest, I first look at long-term results, then always break down by year to confirm.
---
# The Same EA wins in 2022 but loses in 2024
Sometimes there are even more extreme cases with EA development.
With a certain logic,
2022:
- PF: 1.26
- Profit: positive
but
2024:
- PF: 0.87
- Profit: negative
resulted.
The logic and parameters were the same.
Yet the results can differ greatly.
This is because the market itself has changed.
Some years see trends form more easily, others have longer ranges.
Volatility also changes.
Therefore, you must look not only at how it performed over the past five years, but also
**which market conditions it is strong in and which it is weak in**
.
## 5. Profit and Loss Curve
Finally, I always look at the
**Balance / Equity Curve**
.
I may spend more time looking at this than PF.
Ideally, the curve should be
gradually, yet surely, rising to the right.
On the other hand, EAs that
- stay flat for a long time
- suddenly surge in profits
- drop capital rapidly
- concentrate profits in only certain periods
are to be cautioned.
In particular, be careful with EAs that show
“little profit for most of the time, then a few trades yield large profits.”
PF may look excellent in such cases.
---
# Not chasing the “best parameters”
In EA development, there is something I’ve been especially conscious of lately.
That is,
**not chasing the best backtest result**
.
When you optimize, you may end up with
PF1.30
PF1.40
PF1.60
PF2.00
and so on.
You naturally want to pick the parameter with the best number.
However, that parameter may have only matched past moves by coincidence.
In other words,
**curve fitting (overfitting)**
Therefore now I look for a region where the results do not crumble even if numbers change a bit.
For example,
a parameter changes from
20 → PF1.34
21 → PF1.38
22 → PF1.36
23 → PF1.35
24 → PF1.33
If so,
you can feel fairly secure.
On the other hand,
20 → PF1.05
21 → PF1.08
22 → PF2.10
23 → PF1.03
24 → PF0.98
In that case,
PF 22 may be fitting by chance alone.
I personally would not adopt the latter.
---
# Entry conditions vs. conditions not to enter
When I started making EAs,
I used to think about, “What conditions should trigger an entry?”
But now I emphasize
For example,
- ADX is low
- EMA slope is weak
- Price is far from the EMA
- Volatility is extremely low
- Specific time windows
- Friday
- Monday morning
, etc.
Adding such filters reduces the number of trades.
In exchange, you can
avoid making bad trades.
In EAs,
reducing losing trades often improves results more than increasing winning trades.
---
# Is PF1.37 a good number?
So,
**PF1.37**
is that a good number?
My answer is
**“PF alone cannot determine it.”**
.
For example, if you meet these conditions:
- Trades > 300
- Profitable over multiple years
- Maximum DD within tolerance
- No extreme yearly bias
- Profit curve relatively stable
then PF1.37 is highly favorable.
On the contrary, if you have
- 20 trades
- Only one year tested
- Maximum DD 50%
- Profits from a few big wins
then operating in real life would be very cautious.
---
# Summary
When looking at backtests for an EA, PF tends to catch the eye.
However, after validating over a long period, PF is only one of the numbers to evaluate.
What I currently check are the five: **PF, number of trades, maximum drawdown, yearly performance, and profit and loss curve**.
Moreover, I focus on a logic where the results do not crumble significantly even if some parameters change, rather than chasing the set of parameters that yields the best results.
In EA development, it is important to consider not only how much profit you can make, but also how to avoid large losses as much as possible.
In the future, I plan to write about what I learn while actually creating EAs, such as
- where to look during optimization
- how you use ADX and ATR
- the reasons why performance changes year by year
- how to think about the required margin for EAs
- handling trend vs. range markets
I will continue to share what I learn by actually creating EAs.
Note: This article is based on the author's own EA development and backtesting experience. It does not guarantee future profits or performance. Actual trading should be done at your own judgment and responsibility.