The mean reversion thesis is one of the oldest ideas in quantitative finance: assets that have fallen the most tend to recover, and assets that have risen the most tend to give back gains. This is the conceptual foundation of value investing, pairs trading, and dozens of systematic strategies.
The specific version we tested: assets in the bottom 10% of the universe by 7-day return will outperform the top 10% over the following 7 days. Short the leaders, long the laggards.
We call this the Percentile Rank Reversal strategy. We tested it carefully — 40 backtests across 4 distinct market regimes and 10 parameter sets. Here is the honest assessment of what the data shows.
The Testing Framework
Our research process requires a structured evaluation before any thesis receives significant development investment. For the Percentile Rank Reversal, we designed the following framework:
10 parameter sets organized around three variables: the percentile cutoff (how extreme must the underperformance be?), and the holding period (how long to hold the reversion position?).
| Parameter Set | Bottom % | Top % | Hold Days | Rationale |
|---|---|---|---|---|
| 1 | 5 | 95 | 3 | Fastest reversions at extremes |
| 2 | 5 | 95 | 7 | Standard extreme, weekly hold |
| 3 | 5 | 95 | 14 | Extreme entry, patient hold |
| 4 | 10 | 90 | 3 | Moderate cutoff, fast exit |
| 5 | 10 | 90 | 7 | Best theoretical case |
| 6 | 10 | 90 | 14 | Moderate, patient |
| 7 | 20 | 80 | 3 | Conservative cutoff, fast |
| 8 | 20 | 80 | 7 | Conservative, weekly |
| 9 | 15 | 85 | 5 | Asymmetric, favor shorts |
| 10 | 25 | 75 | 5 | Wide net, medium hold |
4 market regime periods chosen to test consistency across different conditions:
| Period | Dates | Regime Type |
|---|---|---|
| Period 1 | 2023 full year | Trending bull |
| Period 2 | Q1-Q2 2024 | Choppy/sideways |
| Period 3 | Q3-Q4 2024 | Secondary rally |
| Period 4 | 2025 YTD | Current market |
A viable strategy must achieve Sharpe above 1.5 in at least three of the four periods. A strategy that only works in one period is not robust — it is regime-dependent.
What We Found: The Short Version
The strategy shows conditional edge. It works in specific regimes and fails in others, and no parameterization achieves consistent Sharpe above 1.5 across three or more of the four test periods.
In choppy/ranging markets (Period 2, Q1-Q2 2024): Every parameter set was positive. The best configurations achieved Sharpe ratios between 1.1 and 1.8, with parameter sets 4 and 5 (10/90 percentile, 3 and 7 day hold) performing best. Win rates ranged from 52% to 61%. This is exactly what mean reversion theory predicts — in ranging markets, momentum reverts and the laggards catch up.
In trending bull markets (Period 1, 2023): Most parameter sets were negative. The strategy was fighting the trend — systematically going long assets that were underperforming in a bull market and shorting assets that were leading. The laggards in a strong uptrend often continue to lag; the leaders often continue to lead. Period 1 was the clearest example of where the reversal thesis breaks down.
In secondary rally (Period 3, Q3-Q4 2024): Mixed results. Some parameter sets were mildly positive, most were near zero or slightly negative. The regime was too directional for pure mean reversion but not strongly trending enough for the long side to dominate.
In current market (Period 4, 2025): Mildly positive for the conservative parameterizations (20/80, parameter sets 7 and 8), near zero or slightly negative for the extreme parameterizations.
The Regime Dependency Problem
The core problem with the Percentile Rank Reversal is that its performance is directly tied to the BTC trend regime. When BTC is trending strongly, the strategy fights the trend. When BTC is ranging, the strategy works.
This means the strategy is not a standalone edge — it is a conditional edge that requires regime detection to be deployed safely. A naively deployed reversal strategy in a strong bull market will systematically short the leaders (which continue to lead) and long the laggards (which continue to lag).
We tested adding a BTC regime filter (similar to the Quality Momentum gate): only deploy the reversal strategy when BTC's 14-day return is within ±10% (ranging conditions) and BTC volatility is above median (active but not panicking). This filter improved results significantly in isolation but introduced a secondary problem: the regime filter's own reliability is imperfect, and a filter that correctly identifies 70% of ranging periods still deploys the strategy in 30% of trending periods where it loses.
The Relationship to the Consolidation Breakout
An interesting finding emerged when we compared the Percentile Rank Reversal to our Consolidation Breakout strategy. The two strategies are directional inverses in certain conditions — the consolidation breakout goes long the laggards (bottom movers by z-score) after a confirmed BTC upside breakout, while the percentile rank reversal goes long the laggards (bottom decile by 7-day return) without requiring a breakout trigger.
The critical difference is the entry confirmation. The consolidation breakout requires a confirmed BTC move before entering. The percentile rank reversal enters based purely on relative performance ranking, with no confirmation that the laggard is actually about to reverse.
This comparison reveals what the consolidation breakout's edge actually is: not the identification of laggards, but the timing of entry after a confirmed catalytic event. The laggard identification is the same. The breakout confirmation is what makes it profitable.
Why We Are Not Retiring This Thesis
The standard conclusion from these results would be: strategy fails viability filter, retire and move to the next thesis. But we are keeping the Percentile Rank Reversal in the active research library, with modifications.
Regime-conditioned version. We are developing a version that uses the BTC regime gate from Quality Momentum as a prerequisite. The reversal strategy only activates during explicitly identified ranging regimes. In trending regimes, the strategy holds zero positions. The preliminary results of this combined approach are more promising — it effectively restricts the reversal strategy to the 30-40% of market time when it has genuine edge.
Sector-relative version. Instead of ranking all assets together, rank assets within sectors (L1s, DeFi, AI tokens, etc.) and long the laggard within each sector while shorting the leader. This reduces the market-level trend exposure because the long and short legs within each sector are more correlated to each other than to the overall market.
Integration with funding rate data. A laggard with negative funding rate (shorts paying longs) is in a fundamentally different position than a laggard with positive funding rate (longs paying shorts). The negative-funding laggard has an additional structural support for a reversion; the positive-funding laggard does not. Adding this filter may improve precision in identifying reversions that are mechanically supported.
The Honest Verdict
The Percentile Rank Reversal, in its basic form, does not meet our viability threshold for deployment. It has a real effect in ranging markets but lacks the robustness to generate consistent positive returns across market regimes.
The positive finding from this research is that it clarifies the mechanism behind other strategies that do work. The consolidation breakout's edge is not primarily in its mean-reversion component — it is in the breakout timing. Quality Momentum's edge is not primarily in the selection of low-volatility assets — it is in the regime gate that keeps it out of the market when the reversal effect dominates momentum.
Understanding why a strategy fails is often as valuable as discovering why one succeeds.
Takeaways
- Mean reversion on short timeframes is real but regime-dependent — it works in ranging markets and fails in trending ones, which limits its standalone utility
- Sharpe ratios of 1.1–1.8 in choppy regimes but negative in trending regimes fail the cross-regime consistency requirement for deployment
- The comparison with the consolidation breakout reveals the importance of entry timing: identifying laggards is easy, knowing when they will reverse requires a confirmation trigger
- A regime-conditioned version (only activate in BTC ranging conditions) and sector-relative ranking are the two most promising modifications currently under development
- Understanding why a strategy fails often clarifies the mechanism behind strategies that succeed — the reversal thesis helped us understand what actually drives the consolidation breakout's edge
- Keeping rejected theses in an active research library is valuable — the right market regime or the right modification may make a currently-unviable thesis deployable