Uncovering Hidden Treasures: Our Deep Dive for Separate Profit Sources

A beginner-friendly summary of the verification: “Uncovering Hidden Treasures: Our Deep Dive for Separate Profit Sources”.

Breakout entry example (XAUUSD daily, real data): buy when price breaks above the recent high.

It’s always exciting to hunt for new edges in algorithmic trading. We’re constantly looking for “alternative” or “separate” profit sources – things that can add a little extra juice to our strategies, especially to hit a consistent 2% monthly return with a safe drawdown. Think of it like trying to find small, independent streams that can feed into a larger river of profit.

What’s the idea?

Our main goal here was to thoroughly explore the FX market for these “separate sources” of profit. We wanted to see if there were hidden patterns, seasonal quirks, or other subtle behaviors in price data that we could exploit. The dream is to find something uncorrelated to our main trend-following strategies, allowing us to stack up small, robust profits for a more diversified and stable overall system. We’re aiming for that sweet spot of around 2% monthly profit without taking on excessive risk.

How I tested it

To make sure our findings were robust and not just random luck, we used a rigorous approach. We didn’t just look at past data and find patterns that might have worked; we used forward testing. This is crucial: it means we tested strategies on data they hadn’t seen before, simulating how they would perform in live trading. It’s our way of avoiding “data dredging,” where you accidentally find patterns that only exist in historical data and vanish in the real world. We cast a wide net, exploring many different mechanisms within price data. This included looking for:

• Calendar effects: Do specific months or days of the week show consistent movements?
• Seasonality: Are there recurring yearly patterns for certain currency pairs?
• Lead-lag relationships: Does one currency pair consistently move before another, or does a pair show predictable self-lag (autocorrelation)?
• Other patterns: Arbitrage opportunities, intraday movements, range reversals, shorting biases, and cross-sectional momentum. We used specific scripts, like scripts/study_calendar_sleeves.py, scripts/study_seasonal_forward.py, and scripts/study_leadlag.py, to systematically test these ideas.

What happened?

Our deep dive yielded some interesting, if not always lucrative, results.

Calendar Effects (Sleeves)

We looked into whether specific months or days offered a consistent edge:

• April & December: These two months did show a slight positive edge. When combined, they offered a +12% return over the test period with a -4.9% drawdown. Even better, they had a low correlation (0.24) to our other strategies, meaning they were somewhat independent. This is great news for diversification!
• Month-end/start: We also confirmed a positive edge around the turn of the month, which is a known phenomenon.
• Wednesdays: Unfortunately, testing for Wednesday-specific effects didn’t pan out. The strategy couldn’t overcome trading costs due to its high frequency. In other words, the profit per trade was so tiny that commissions and spreads ate it all up. Takeaway: Only low-frequency calendar effects (like specific months or month-ends) have enough resilience to cover trading costs. High-frequency calendar patterns are generally not profitable.

Currency-Specific Monthly Seasonality

We forward-tested if particular currency pairs exhibited reliable seasonal patterns throughout the year.

• The result: This strategy generated a mere +1.7% over 8 years, winning in only 4 out of 9 years, with a correlation of 0.33.
• Conclusion: This was simply too weak and inconsistent to be considered a viable “separate source.” It was rejected.

Lead-Lag Discovery

We investigated if one currency pair consistently “led” another, or if a pair consistently “lagged” itself (meaning its past movements predicted its future).

• Between pairs: We found zero instances of persistent lead-lag relationships between different currency pairs. Nada.
• Self-lag: For a pair predicting its own future (autocorrelation), many showed a tiny negative edge (-0.02% to -0.04%) at H1 (1-hour) intervals. This indicated a very weak reversal tendency.
• Conclusion: This was also rejected. The edge was minuscule and required high-frequency trading, meaning it would be completely wiped out by trading costs.

What I learned: The Big Picture

After thoroughly exploring numerous mechanisms and rigorously forward-testing them, we’ve reached a pretty firm conclusion: The “real” separate profit sources within price data are extremely small.

• Only low-frequency calendar effects (like month-end/start, April, and December) proved to be genuine and robust.
• Each of these contributes a tiny amount, roughly ~0.5% per year.
• Even if we stack all these minor calendar effects together, we’re looking at maybe ~1% per year. While positive, that’s a long way from our target of 2% monthly profit! Crucially, we found no significant “separate sources” within price data itself. This isn’t a hasty conclusion; it’s based on extensive testing across many different mechanisms, all validated with forward testing. We’ve gone as deep as we can without falling into the trap of “data dredging” (finding patterns that only exist because we looked hard enough in the past).

The Path to 2% Monthly Profit (with Safe Drawdown)

Our extensive research solidifies that there are two primary, realistic paths to consistently achieve our 2% monthly profit goal with safe drawdown:

1. Refining our core trend-following strategy: The only truly viable path within price data is to further sharpen and optimize our existing core trend-following strategy. For example, our fto breakout_h1 strategy, when applied across 7 currency pairs with an overlay system, has demonstrated an impressive ~2.62% monthly return with a manageable 9.4% drawdown. This is the gold standard for what’s achievable within price data.
2. Leveraging external (non-price) data: The most promising avenue for finding truly significant and independent profit streams lies outside of raw price data. Think about factors like carry trade strategies (profiting from interest rate differentials between currencies). This is where the big, uncorrelated edges are likely to be found. It’s worth noting that we’re already making good progress on our prop firm qualification, having achieved about 79% using a diversified system. This journey of discovery continues, and while the “alternative sources” within price data might be tiny, understanding their limitations helps us focus our efforts on the most promising avenues!

Code to reproduce

You can reproduce this with the following scripts (see repo).

• scripts/study_calendar_sleeves.py
• scripts/study_leadlag.py
• scripts/study_seasonal_forward.py