Does volume profile survive walk-forward testing?

Walk-forward testing: decide the rules on the past, then test on unseen future data (no hindsight).

Walk-forward testing: decide the rules on the past, then test on unseen future data (no hindsight).

Integrating volume profile data into my existing trading framework has yielded clear results, confirming that while volume analysis is useful, it does not outperform established swing-level strategies. I took the POC (Point of Control) and Value Area data from my previous research and integrated them into my proven system, which uses Dow theory, multi-timeframe analysis, and pull-back entries. I then ran a forward test to compare this against the swing-level strategy I established in research 73.

Comparison of Filter Performance

Strategy	Performance (Years Profitable)	Return (%)
Swing-levels (baseline)	5 / 6	+42.2
Volume Profile (vp0.5)	4 / 6	+30.6
Volume Profile (vp1.0)	4 / 6	+16.6
Swing-levels + vp0.5	4 / 6	+49.0
The results show a clear hierarchy. My original swing-level filters remain the “champion” of the strategy, successfully hitting the target in 5 out of 6 years. The volume profile filters, while profitable, failed to meet my benchmark of 5 out of 6 profitable years. In other words, they are reliable enough to keep in the toolkit, but they aren’t the primary drivers of success. Interestingly, combining the two actually caused a slight degradation in consistency, dropping from 5/6 to 4/6.

Next Steps

The takeaway here is that individual filters have reached a point of diminishing returns. Testing channels or price action patterns previously resulted in outright failure or performance degradation, and volume profiles sit somewhere in the middle. They are certainly not “trash,” but they don’t hold the same predictive weight as price-based swing levels. I plan to keep the volume profile engine as a base component, as it may prove more effective for assets where volume is more reflective of actual market depth, such as indices or gold. Moving forward, I am done testing filters in isolation. My next phase is to build a confluence-based scoring system. I will integrate all engines (levels, profiles, pivots, and Fibonacci) into a single metric to see if calculating the overlap of these signals can finally create a more robust “best of” combination. This will serve as the final synthesis of my current research framework.

How this connects

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