14% More Profit? Why This “Aggregated Vol-Target” Idea Was Shelved (Again)

A beginner-friendly summary of the verification: “14% More Profit? Why This “Aggregated Vol-Target” Idea Was Shelved (Again)”.

What’s the Big Idea?

Sometimes, an idea for improving our algorithmic trading (EA) systems seems brilliant on paper, but faces practical roadblocks. That’s exactly what happened with our “aggregated volatility targeting” concept a while back. Here’s the backstory: In an earlier research project (Research 88), we explored a way to manage risk and position sizing across multiple trading strategies (which we often call “sleeves”). Instead of each sleeve adjusting its own position based on its individual volatility, we tried combining all their equity into one big pool. Then, we’d adjust positions for all sleeves based on this aggregated equity’s volatility. The results were impressive: a +14% improvement over the “distributed” approach, where each sleeve acts independently. So, why didn’t we use it? The catch was complexity. Implementing this “aggregated” system required a central “master orchestrator” to manage everything. This created a single point of failure and made the whole setup much more complicated to operate for our users. So, despite the performance boost, we shelved it. Fast forward to today: our deployment method has evolved. We now primarily use a single MT5 Expert Advisor (EA) per trading account. Crucially, this EA already has access to the account’s total equity. This change meant that implementing the aggregated volatility targeting would now be almost free in terms of additional cost or complexity! The original reason for rejection was gone. Given this new reality, it was time for a fresh look. Could this “aggregated” idea finally deliver its promised performance boost with our latest system, v1.4.0 (which includes our new “stock filter” and vt_cap3.0 improvements)? We decided to re-evaluate it with a rigorous, two-pass iterative testing method, ensuring no “hindsight” was used – meaning we only used past market data to make decisions, just like a real EA would.

How I Tested It

To see if aggregated volatility targeting was still worth it, I put it head-to-head against our current v1.4.0 system (which uses the distributed approach). I ran extensive backtests covering a long historical period, making sure to compare apples to apples. The main metrics I focused on were:

Monthly Profit: How much profit the system generated on average each month.
Drawdown (DD): The maximum drop from a peak in equity. A 10% monthly DD means that at its worst, the account dropped 10% from its highest point within a given month. Lower is always better! I wanted to see if the aggregated approach could still provide that juicy performance uplift, especially now that the implementation barrier was virtually non-existent.

What Happened?

This is where things get interesting – and a little disappointing, but in a very educational way!

The Initial Spark: A Glimmer of Hope

At first glance, the aggregated approach looked quite promising over the full testing period. Let’s compare:

Our current Distributed v1.4.0 system: Showed a monthly profit of +1.59% with a 10% maximum monthly drawdown.
The Aggregated system (specifically, iteration L2): Achieved a monthly profit of +1.68%, which is a +6% improvement over the distributed version! This looked good! A +6% boost without much extra cost? Sign us up, right? However, this +6% improvement was significantly less than the +14% we saw in Research 88. Why the difference? It turns out our v1.4.0 system already has a sophisticated “stock filter.” This filter is designed to pick up on broad market risk-off signals (often stemming from the stock market) and adjust trading across all sleeves accordingly. Essentially, our current system was already doing a great job of managing cross-sleeve risk, which reduced the unique benefit of the aggregated volatility targeting. But the real problems emerged when we dug deeper.

The First Hurdle: Unstable Results

Our testing process involves multiple “gates” or checks, and the aggregated system stumbled at the very first one: convergence. Imagine you’re trying to balance a wobbly table by repeatedly adjusting its legs. That’s a bit like what happened here. The aggregated system’s leverage calculation is “self-referential” – meaning the leverage affects the equity, which then affects the leverage, and so on. This created a feedback loop that caused the results to oscillate rather than settle on a stable value. We saw the monthly profit figures for the 10% DD scenarios jump around:

First iteration (L1): +1.61%
Second iteration (L2): +1.68%
Third iteration (L3): +1.50% Notice how it went up, then down significantly? This oscillation, with a difference of 0.134 between L1 and L2, meant the system was unstable. The final result depended on which iteration you chose, which is a big red flag for robustness. You want a system that consistently converges to the same optimal solution. Our distributed system, on the other hand, doesn’t have this problem. Each trading stream manages its own volatility independently, which makes the structure inherently stable and ensures it converges reliably. This was a critical flaw for the aggregated approach.

The Second Hurdle: It Only Works in Specific Conditions

Even if we ignored the instability, the aggregated system hit another major roadblock: regime splitting in sub-periods. Think of it like a sports team that only performs well in certain weather conditions. While the overall numbers might look okay, the aggregated system’s performance wasn’t consistent across different market environments.

Weak Market Period (2015-2020): During this period, the aggregated system did perform better (+0.71% monthly profit) compared to the distributed system (+0.61%). So, in tough times, it seemed to shine.
Strong Market Period (2021-2026, which is closer to our current market regime): Here, the tables turned completely. The distributed system was clearly superior, achieving +2.55% monthly profit compared to the aggregated system’s +2.45%. This means the overall +6% improvement we initially observed was only driven by its outperformance during the weak market period. For the current, stronger market conditions, the aggregated approach would actually be disadvantageous! This is a crucial distinction. We want systems that perform well in the current market and are robust across various conditions, not just specific past regimes.

What I Learned

This research was a fantastic example of why rigorous testing and our multi-stage validation gates are so important. What initially looked like a promising +6% improvement (a “false positive”) was correctly rejected when we applied our full verification discipline. Here’s the bottom line:

Our current Distributed Volatility Targeting (v1.4.0) is robust and correct. It’s already doing an excellent job.
The initial hypothesis that we could get a “free +14%” boost by implementing aggregated volatility targeting with a single EA was debunked. The v1.4.0 system’s “stock filter” already provides similar benefits, making the aggregated approach largely redundant.
The original decision in Research 88 to not adopt aggregated volatility targeting was, in fact, correct – not just because of operational complexity, but also due to its lack of robustness, as proven by its instability and regime-dependent performance. So, for now, our confirmed system remains unchanged. This experiment reinforced that sometimes, the simpler, more stable approach is indeed the best, and our thorough testing process successfully prevented us from implementing a potentially unreliable change. On to the next research!

How this connects

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

Beyond the Limits: How Volatility Targeting Unlocked Real FX…