What happens to the guardian when you avoid indicators? Confirmed with 12 years of backtesting

Serial: Yoko and Kenji's GOLD EA Mastery Notebook Part 9

Hello, I’m Yoko.

In the past few days, Kenji at my place has been a little restless. Even during dinner, with a smartphone in hand, he comments live on the progress of back-testing: “Ah, we’ve passed 2014,” “Look, around 2019,” which makes the meal go cold.

The reason is,I loaded the economic indicator calendar as CSV and can verify the effect of the indicator filter in backteststhis time. I will organize and share the results of that “excitement.”

What became possible

In MT5 backtest (Strategy Tester), the built-in economic calendar cannot be used. Therefore, until now, we could not verify the effect of the guardian’s “wait before and after indicators” feature in backtests.

So Kenji wrote the MT5 calendar data to a CSV file and created a system to read it during backtests. This enables validation based on the past 12 years of indicators.

The validation period is 12 years from July 2014 to July 2026, with fixed lot 0.50, starting balance 1,000,000 yen, and a low-spread account environment. We varied only the indicator-related settings and ran guardian in many variations.

First finding: the initial indicator filter setting didn’t work

We were surprised at first. In the initial guardian setting (indicator filter ON, no orders deleted), the result was exactly the same as with no filter. Over 12 years, it never worked.It was completely the same. It hadn’t worked even once in 12 years.

The reason was a time offset. Guardian places orders around 18:00 Japan time in the evening. However, important US indicators often come out around 21:30–22:30 Japan time, so at the moment the orders were placed, they had not yet entered the indicator’s window. The initial filter only acted to skip when the orders overlapped with the indicator at the moment of placement.

Subsequent indicators did not protect already placed orders or positions that were filled.

Thus, two settings were tested

The “restore” in ② is newly created this time. Kenji thought, “if you leave it deleted, you might miss the price movement that starts after the indicator.”

The results were as follows

Note: “High” and “Mid” are the classification of economic indicator importance. The recalculated figures differ slightly from the numbers on the sales page.

From this table, three things are clear.

Even with D (avoiding both high and mid), PF was 2.39 and 12-year net profit about 4.18 million yen.Even without relying on indicators, the very idea of “aiming for a breakout of the previous day’s range” has a certain edge, so it’s not just a simple safety mechanism.

How often did the “back-and-forth slap” occur?

The scariest moment with indicators is when both buy and sell orders fill, and both get stopped out—the “back-and-forth slap.” Kenji checked all 12 years of trades.

Days when both buy and sell filled were about 112 days out of ~2,800 trading days (about 4%). Of those, both sides lost on 6 days. Moreover, the instantaneous back-and-forth where both filled within minutes to tens of minutes occurred on 8 days in 12 years, and only on 1 day did both lose. The total of the 112 days with both orders filled was actually positive, because on days when a trend appears, both buy and sell can extend in sequence.

So, how to apply it

In response to these results, Kenji is hurriedly preparing a new version that includes the Restore function. The settings will be a bit more, but we will write them clearly in the specification. Here are our usage suggestions.

In particular, for those who manage Guardian across multiple accounts, you can apply different types per account. For example, make the first account Standard, the second Balanced, etc. It’s a risk diversification approach to ensure not all accounts yield the same results on the same day.

Yoko’s closing remark

Kenji believed that “avoiding indicators would make it safer.” On a night when the numbers clearly betrayed him, after a moment of silence he returned to his computer and asked, “Then maybe restore is enough.” The dinner grew cold again.

Even when assumptions are overturned, we verify, adjust, and re-verify. This iterative process will gradually make this EA more reliable. We will continue to share both the good and the not-so-good findings.

Next issue preview

Next time is the final part of the series. We will honestly discuss the current state of the Golden Guardian. We will also share the latest progress of our accounts, trading on accounts with wider spreads, and a recap of everything so far.

Thank you for reading until the end.

Yoko

※This article presents past validation results and does not guarantee future profits. Please make investment decisions at your own risk.