Yosuga EA's Secret: Limiting Entries to "Line Vicinity" Boosts Performance!

A beginner-friendly summary of the verification: “Yosuga EA’s Secret: Limiting Entries to “Line Vicinity” Boosts Performance!”.

This post wraps up our deep dive into the ‘Yosuga-style’ trading method, specifically testing a key discretionary filter. The goal was to see if a particular rule, claimed to boost profitability, could be successfully turned into a mechanical Expert Advisor (EA).

What’s the idea?

Yosuga’s method, as taught in their materials, includes a specific “lever” for boosting trading performance: only taking trades when a trend reversal happens near a significant price level. This specific rule was said to boost overall profitability, potentially from 62% to 71% (likely referring to a win rate or a similar performance metric). My mission was to see if this powerful-sounding discretionary rule could be turned into a fully mechanical Expert Advisor (EA) that consistently performs in the real world.

How I tested it

To mechanize this concept, I focused on identifying those “prominent highs/lows.” I used what’s called “Dow structure” – a way to define significant swing points in the market – to draw horizontal lines at these key levels. Then, I only allowed the EA to enter a trade when the price was within a certain distance of one of these lines. This distance was measured using Average True Range (ATR). ATR (Average True Range) is a common way to measure market volatility – essentially, how much the price moves on average over a given period. Using it helps us define a dynamic “nearness” rather than a fixed number of pips. I tested four different scenarios to see which, if any, performed best:

OFF: No filter applied – the base strategy.
Line 0.5 ATR: Entries only if the price was within 0.5 times the ATR of a significant line.
Line 1.0 ATR: Entries only if the price was within 1.0 times the ATR of a significant line.
Line 1.5 ATR: Entries only if the price was within 1.5 times the ATR of a significant line. Crucially, I used “forward testing” for this comparison. This means testing the strategy on data it’s never seen before, which is the only real way to know if it’s genuinely robust and not just lucky on past data (what we call ‘in-sample’ optimization). Our key metric, as always, was the Profit Factor (PF), which is your gross profit divided by your gross loss. Anything above 1 means you’re profitable overall.

What happened?

The results were a bit of a reality check. Initially, on the “in-sample” data (the historical data used for development), the filter did reduce the number of trades and slightly improved the PF from 1.05 to 1.11. Sounds promising, right? However, when we moved to the unseen “forward testing” period, the picture changed dramatically. The filter did not prove to be robust across the board. Here’s how the different versions performed over their respective forward testing periods:

OFF (no filter): +26.4% profit over months 3-6
Line 1.0 ATR filter: +4.2% profit over months 4-6
Line 1.5 ATR filter: +1.1% profit over months 4-6 In other words, the strategy without the special filter performed significantly better. None of the filtered versions came close to hitting our target of profitability in at least 5 out of 6 months. In fact, applying the filter actually shrunk overall performance, and it didn’t fix the strategy’s negative performance in 2019. It seems the filter just reduced the number of trades without actually adding a robust “edge” – a statistical advantage – to the strategy.

What I learned: The Limits of Mechanization

This study delivered some really important insights into the nature of profitable trading, especially when trying to bridge the gap between discretionary (human judgment) and mechanical (EA) systems.

The Core Logic is Sound: First, the good news: the fundamental, trend-following logic at the heart of Yosuga’s method can be successfully mechanized. This confirms what I’ve found in my own research – there’s a genuine, mechanical “edge” there. So, the core idea is solid! ✅
The “Extra Sauce” Remains Discretionary: However, the specific “lever” that was supposed to provide the extra boost in profitability (the jump from 62% to 76% in the original material’s claims) could not be successfully translated into a robust, forward-testable mechanical rule. ❌ This suggests that the “alpha” – the excess profit – in Yosuga’s method comes from elements that are inherently discretionary and context-dependent. These likely include:

Choosing the ‘right’ lines: The process of drawing those significant high/low lines (even if it’s a 6-step process) involves human judgment about which ones are truly effective at any given time.
Reading price action: Interpreting how price interacts with those lines – what we call ‘price action’ – is a nuanced skill that’s hard to put into rigid code.
Trade management and scenario planning: Deciding how to manage an open trade, adapting to changing market conditions, and planning for different outcomes are all highly subjective. This finding aligns perfectly with what we discovered in Research 33, where we tried (and failed) to use machine learning to “meta-label” and select winning trades. It seems that truly discretionary trade selection, whether through human rules or even advanced AI, is incredibly difficult to mechanize into simple, robust rules. It’s a similar pattern to what we saw with the “Ishinnokai” method in Research 54: the core trading concept is valid, but the significant gains come from the trader’s discretionary skill, not just the mechanical rules. While it’s philosophically appealing to believe that “discretion is ultimately logical,” this experiment shows that reducing that logic to simple, forward-testable mechanical rules doesn’t always work, even with genuine educational material. Attempting to mechanize every last bit of discretionary insight often leads to “data mining” – finding patterns that only exist in past data and won’t hold up in the future. For now, the “Yosuga-style” approach, in terms of its additional filters, is a closed chapter for our mechanical system development. Our Core System v1.1.0 remains the most robust and effective version we have.

How this connects

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