Hunting exhaustively for a second source of returns

Adding extra trading strategies to a portfolio to boost performance is a common goal, but my latest research suggests that finding additional sources of profit within price data is nearly impossible. After testing a wide variety of mechanisms, including seasonal patterns, lead-lag relationships, and mean reversion, it is clear that the “holy grail” of extra alpha simply does not exist in the charts.

The search for extra alpha

I spent significant time exploring different ways to stack strategies. My findings for these specific calendar-based filters are summarized below:

Strategy	Performance	Verdict
April & December Filter	+12% Total / 4.9% DD / 0.24 Correlation	Adopted
Month-End/Start	Positive, but limited	Adopted
Wednesday Trading	High frequency, cost-inefficient	Rejected
Seasonal Forward Testing	+1.7% / 8 years (4/9 years profitable)	Rejected
The April and December filters, along with month-end transitions, were the only ones that showed genuine promise. However, their impact is tiny, contributing only about 0.5% per year each. Even if you stacked every viable calendar filter I identified, you would only reach an annual return of about 1%, which falls far short of a 2% monthly target.
I also investigated lead-lag relationships (the idea that one asset moves before another) and self-correlation. I found zero instances of persistent lead-lag behavior. Self-correlation was slightly negative (-0.02 to -0.04), indicating a weak mean-reverting tendency on H1 (one-hour) timeframes. Because these signals occur frequently, the costs of trading them quickly eat the profits. In other words, the strategy fails because the transaction costs are higher than the tiny edge it provides.

The two paths forward

After testing everything from arbitrage and seasonality to cross-sectional momentum, I am confident that there are no “hidden” major sources of profit within price data. Any further searching within the same data is likely to be data-dredging, where you find patterns that look real but are actually just statistical noise. If the goal is to achieve a 2% monthly return while maintaining a safe drawdown, there are only two realistic paths:

Refining the Trend Core: The only way to improve performance using price data is to sharpen the core trend-following engine. My separate research track (using 7 pairs with an overlay) has already reached a level of 2.62% monthly return with a 9.4% drawdown. This is the only proven path within the charts.
Using External Data: The other option is to look beyond price data entirely, such as using carry trade or interest rate differentials. While this conclusion might seem restrictive, it is the result of rigorous forward testing across multiple mechanisms. I have already achieved a 79% success rate in passing prop-firm evaluations using a diversified system, so the focus now shifts from “finding” new edges to refining the ones that actually work.

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