Filtering trends by how information arrives: also no

The “frog-in-the-pan” hypothesis suggests that markets underreact to information that is released gradually. This means trends built on a series of small, consistent price moves are more sustainable than those driven by sudden, large jumps. I recently tested this concept to see if filtering for “price continuity” could improve momentum trading, but the results were clear: it does not add value to the strategy.

The hypothesis and test design

Most momentum strategies rely on the magnitude of price changes, such as how far a price has moved over a set period. My hypothesis was that the way a price moves matters more than the distance. By focusing on the ratio of positive versus negative bars, I could filter for “stealth” trends: price action characterized by a high frequency of small, same-direction bars rather than erratic gaps. I ran this test across 20 instruments (including FX, metals, and stock indices) using a lookback period of 60, 120, and 240 days. I compared three groups:

Plain: Standard momentum (all signals).
Cont: Continuous signals (filtered for “stealth” trends).
Disc: Discrete signals (the inverse, filtered for large, jumpy trends).

Why it failed

The performance was disappointing across the board. The fundamental issue is that the underlying momentum strategy failed to show an edge even before applying the filter. In this universe of assets, naive directional momentum has been consistently ineffective, with OOS (out-of-sample) Sharpe ratios ranging between -0.1 and -0.8. In other words, because the raw momentum signals lack predictive power, filtering them for “continuity” is essentially just refining noise. The best-performing “continuous” configuration still yielded an OOS return of -3.0% with a -9.95% drawdown and a 7.9% MC (Monte Carlo pass rate, or the probability of meeting the performance requirements of a prop firm).

The takeaway

This experiment confirms that “continuity” is not the missing link for momentum in these markets. When the core strategy lacks an edge, applying sophisticated filters to the inputs does not create one. It is a reminder that you cannot refine a zero-edge signal into a profitable one. Moving forward, I have two other price-based ideas to explore: Garman-Klass path efficiency and intraday volatility standardization. However, both will be tested as overlays to existing trend-following cores rather than as standalone entry signals. Given the difficulty of finding new directional edges in price data alone, my next research track will likely shift toward dynamic risk scaling, focusing on managing position sizes based on the distance to equity knock-out levels rather than trying to predict market direction.

How this connects

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