FX Backtest Diary
日本語 English

Yosuga's Dow Method: Can This Trend-Following Secret Be Automated by EA?

✍ etherpoc Method verification · 6 min

## What's the Idea Behind This EA?

A beginner-friendly summary of the verification: “Yosuga’s Dow Method: Can This Trend-Following Secret Be Automated by EA?”.

What’s the Idea Behind This EA?

We recently got our hands on some intriguing educational materials from a trader known as Yosuga (dating back to 2022). His core philosophy really resonated with us: “Trend following is the only way to win in FX,” and “FX is all about catching the very beginning of a trend.” Guess what? This aligns perfectly with what we’ve discovered in our own research here on the blog! It’s always great when independent conclusions point in the same direction. So, the goal for this project was to take Yosuga’s Dow Theory-based trend trading strategy and see if we could turn it into an Expert Advisor (EA) – basically, automate it. Could we capture his successful insights in a machine-driven trading system?

How We Mechanized and Tested It

To mechanize Yosuga’s approach, we focused on its core logic. Here’s how we broke it down:

  1. Defining Trend Reversals: We used a “large zigzag” indicator (let’s call it Z1, which focuses on deeper price swings) to define when a Dow Theory trend reversal occurred. Think of it like identifying the major shifts in the market’s direction.
  2. Multi-Timeframe Confirmation: We then checked if this trend direction aligned with what was happening on a Higher Time Frame (HTF) – this is often called Multi-Timeframe (MTF) analysis. It’s like checking the bigger picture to confirm the local trend.
  3. Entry Strategy: The “Pullback”: Once a trend reversal was confirmed, the EA would wait for a “pullback” – a temporary dip or rise against the new trend. The entry signal fired when a “small zigzag” (Z2, catching minor swings) broke past the previous swing in the new trend direction. This is designed to catch the “beginning of a trend” after a brief correction, often called an “entry on a pullback after reversal.”
  4. Stop Loss and Exit: The EA would hold the trade until another trend reversal occurred, specifically if the swing low (for a long trade) or swing high (for a short trade) that defined the current trend was broken. We approximated the zigzag indicator using what we call dow_structure – a fractal-based method that reliably identifies confirmed swings without any “look-ahead bias” (meaning it doesn’t cheat by using future data). Now, here’s a crucial point: Yosuga’s original material included some additional “leverage” elements – things like limiting trades to specific price line vicinities, incorporating price action analysis, and making scenario-based decisions. These are quite discretionary, meaning they rely on a trader’s judgment. For this initial test, we only mechanized the core 62% of his strategy, leaving out these more subjective elements. We wanted to see if the fundamental trend-following logic itself held up.

So, What Happened? The Test Results

We put our mechanized Yosuga EA through its paces in a multi-step verification process.

Step 1: Initial Backtest (H1/H4 Long Trades)

We ran an initial backtest focusing on long trades on the H1 (1-hour) and H4 (4-hour) timeframes. The results were:

  • Profit: +6.6%
  • Drawdown (DD): -17.6%
  • Profit Factor (PF): 1.05
  • Winning Years: 3 out of 7 Let’s break that down:
  • Drawdown (DD) is the largest peak-to-trough decline in your account balance during a period. A -17.6% DD means your account dropped by almost a fifth at its worst point. That’s a pretty significant risk for a relatively small +6.6% gain.
  • Profit Factor (PF) is calculated as your gross profit divided by your gross loss. A PF of 1.05 means for every $1 you lost, you made $1.05. While it’s above 1 (meaning profitable), it’s just barely profitable.
  • Winning Years: Only 3 out of 7 years showed a profit. This tells us the strategy wasn’t consistently profitable year after year. In other words, this initial test showed a small profit, but with considerable risk and not a very consistent performance. It wasn’t a home run.

Step 2: Correlation with Trend Core

Next, we looked at how well our mechanized logic actually correlated with real market trends. We found a correlation of +0.65 with what we define as the “trend core.” This is a good sign! It means the EA’s logic genuinely is tracking trend behavior. It confirms that the underlying idea of identifying and following trends, as Yosuga proposes, is indeed present in our EA.

Step 3: Full Forward Test

Finally, we ran a comprehensive forward test, which simulates how the EA would perform on new, unseen data, giving us a more realistic picture of its potential.

  • Total Profit: +28.1%
  • Winning Years: 3 out of 6 (meaning it met our internal standard of 5 profitable years out of 6) – △ Weak
  • Key Observation: A large portion of that +28.1% profit was heavily dependent on performance in 2020. The Drawdown (DD) remained significant. So, while the overall profit figure looks decent on paper, the consistency wasn’t there. It didn’t meet our standard for robust performance, and the reliance on a single strong year, coupled with large drawdowns, means it’s not a truly reliable performer yet. It performed better than some previous strategies we’ve tested (like “Ishin-no-kai,” which failed its forward test), suggesting that the discipline of Dow Theory combined with Multi-Timeframe analysis does have some merit. However, it was quite similar to other “pullback” type strategies we’ve explored (Research #42 and #54), which also struggled with consistent forward performance.

What We Learned: The Takeaways

This research project gave us some really valuable insights into automating discretionary trading strategies:

  1. The Core Logic is Valid (and Mechanizable): Yosuga’s fundamental idea that “trend following is the only way to win” and that FX is about “catching the beginning of a trend” is sound. We were able to mechanize this core logic, and it does indeed track trends (as shown by the +0.65 correlation).
  2. Pure Mechanization Has Limits: While the core logic is genuine, its purely automated form didn’t outperform our existing core strategies or provide a consistently robust edge in forward testing. The drawdowns were still too high, and performance was too inconsistent across different years.
  3. The “Secret Sauce” is Likely in the Discretionary Parts: Yosuga reportedly earns a significant income (around 40 million JPY annually). Our analysis strongly suggests that the bulk of his success probably comes from the “leverage” elements we didn’t mechanize – things like precise line selection, nuanced price action interpretation, and advanced scenario analysis. These are the highly discretionary parts that can boost a 62% core strategy up to 76% or more. This highlights the challenge of translating human intuition and experience into rigid algorithms.
  4. Discretionary ≠ Un-mechanizable (Always): The good news is that this study partially proved our hypothesis that “discretionary logic can be mechanized” – at least for the core trend-following idea. The next frontier is to see if we can mechanize those “additional 14%” discretionary elements into a robust, forward-tested edge. What’s Next? Our immediate next step is to try and mechanize one of those powerful “leverage” filters: the “line vicinity limit.” We’ll add this filter to our existing EA and run another forward test to see if it significantly improves consistency and reduces drawdown. For now, there are no changes to our established trading systems. We’re always learning, and this experiment was a fantastic step in understanding the nuances of automating successful discretionary strategies!

How this connects

This verification builds on earlier ones (what failed before and what I tried this time, comparisons between approaches).