Can filters rescue short-timeframe trend trading?

Adding filters to a short-term breakout strategy does improve performance metrics, but it is ultimately not enough to overcome the inherent structural weaknesses of lower timeframes. I started this research as a follow-up to my previous work on the BreakoutLong system. My goal was to see if adding “higher timeframe” filters, specifically aligning trades with the trend of a larger timeframe using a non-leaking SMA (Simple Moving Average) calculation and ADX for trend strength, could polish the performance. I first verified that the core system remained stable, yielding an 89.4% return with a 9.0% drawdown and an 85.2% MC (Monte Carlo pass rate, or the probability of passing prop-firm evaluation rules). I tested five different filter combinations across M15 and M30 timeframes. Here is how the most effective configurations performed:

Configuration	Trades	PF	Max DD
M30 (Base)	3009	1.19	-16.2%
M30 + D1 Trend	1882	1.25	-11.5%
M30 + D1 Trend & ADX > 20	1463	1.26	-10.5%
The filters worked exactly as designed. By cutting out low-quality trades, the system reduced its total trade count, which mitigated the impact of transaction costs and helped improve the PF (profit factor, or gross profit divided by gross loss) and DD (drawdown, or the peak-to-trough decline). The D1 timeframe filter was the most effective.
However, even with these improvements, the results highlight a hard limit. While M30 showed promise, the M15 timeframe still failed to sustain a PF above 1.0 when accounting for realistic 0.5 pip transaction costs. The M5 timeframe performed even worse, remaining unviable regardless of the filters applied.
In other words, the filters successfully refined the M30 data, but they could not turn a fundamentally weak timeframe into a strong one. Even my best-filtered M30 system (PF 1.26, DD -10.5%) was still inferior to the raw, unfiltered H1 timeframe system (PF 1.36, DD -8.6%) across all prop-firm metrics.
The takeaway is clear: trying to polish a short-term system with extra filters is less effective than simply using a more stable, higher-timeframe system from the start. I have decided to keep these filter functions as a general feature for the BreakoutLong code, as they may be useful for other research tracks or different timeframes in the future. For the core system, however, I am making no changes. Relying on H1, H4, and D1 timeframes remains the most robust path forward.

How this connects

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