Beyond Raw Returns: How Asymmetry, Skew, and Stop-Loss Survival Reveal Hidden Predictive Signals in Cryptocurrency Scores

2.1 Dataset

2.2 Return Metrics Definitions

2.3 Correlation Framework

For each of the 192 scoring dimensions, we compute Pearson and Spearman rank correlations against each return metric, segmented by four market regimes (bull, mild bull, mild bear, bear) derived from Total3 forward returns. The combined correlation is a weighted average across regimes, proportional to the number of trading days in each regime (bull: 69, mild bull: 86, mild bear: 94, bear: 75 days).

2.4 Walk-Forward Validation

We employ a rolling walk-forward design: 60% of dates form the training window (minimum 90 days), with 30-day out-of-sample test windows sliding forward. This produces 4 non-overlapping test periods. For each scoring dimension and target metric, we report mean OOS Spearman, standard deviation, and % same-sign consistency across windows.

3.1 Baseline Stop-Loss Survival Rates

Before examining predictive signals, we note the base rates for stop-loss survival across the full dataset:

The stop-loss is triggered before the take-profit target in approximately 70–75% of cases. This establishes that cryptocurrency markets exhibit a strong downside-first tendency, making the identification of scores that improve these odds particularly valuable.

3.2 Top Predictive Scores by Metric

Table 1. Strongest Spearman correlations at the 7-day horizon (combined across regimes)

Table 2. Strongest Spearman correlations at the 30-day horizon

3.3 Signal Orthogonality: Each Metric Reveals Different Scores

A central finding of this study is the low overlap between top-ranking scores across metrics. When comparing the top 15 scoring dimensions for each metric at the 7-day horizon:

This demonstrates that each metric captures a genuinely different aspect of price behavior, and that a score's ability to predict raw returns provides essentially no information about its ability to predict return asymmetry or distributional skew.

3.4 Walk-Forward Stability

Table 3. Most temporally stable signals (100% same-sign consistency across OOS windows)

All reported signals maintain their directional sign in every out-of-sample window tested. The low standard deviations — particularly for bug bounty (0.013) and price change 24h (0.028) — indicate remarkably stable effects.

4.1 Open Interest: The Stop-Loss Predictor

Open interest total emerges as the single most powerful predictor of stop-loss survival, achieving Spearman correlations of +0.13 to +0.19 across all stop-hit configurations and both horizons, with 100% directional consistency in walk-forward validation.

Two complementary mechanisms may explain this effect. First, high open interest reflects active derivative markets with deep liquidity, which provides structural support for price recovery after dips. Second — and perhaps more importantly — elevated open interest creates financial incentives for market makers and large participants to engineer short squeezes. When substantial short positions accumulate, pushing the price upward triggers cascading liquidations that are financially profitable for the actors initiating the move.

Notably, open interest ranks only modestly for raw return prediction — it does not tell you how much a cryptocurrency will gain, but rather how the path will unfold. This distinction is precisely what our multi-metric framework captures.

4.2 Momentum Scores: Universal but Differentiated

Price change scores (24h, 7d, 14d) appear across multiple metrics but with varying strengths and horizons. For raw returns, short-term momentum (24h) has modest power (rho = 0.11). For asymmetry and stop-hit, momentum is the dominant signal (rho up to 0.16 for 24h), consistent with the observation that assets in upward trends exhibit favorable risk profiles for trailing stops. For skew, momentum signals are weak or absent, replaced by fundamental metrics.

This suggests that momentum predicts the direction and risk profile of returns, but not the shape of the distribution.

4.3 ATH Change Percentage: The Contrarian Signal

Distance from all-time high proves to be the most stable raw return predictor (walk-forward OOS rho = +0.239, sign = 100%). Cryptocurrencies trading far below their ATH tend to outperform — a classic mean-reversion effect. However, this signal does not translate to stop-loss survival or skew prediction, confirming that "likely to go up" and "unlikely to crash first" are genuinely distinct properties.

4.4 Revenue/Market Cap: The Skew Anomaly

Perhaps the most counterintuitive finding: protocols with high revenue relative to market capitalization exhibit negative forward skew (rho = −0.17, walk-forward sign = 100%). In plain terms, high-revenue protocols experience more extreme downward moves than upward moves over 30-day windows.

The likely explanation is behavioral: these are mature DeFi protocols (Aave, Uniswap, etc.) that attract profit-taking. Their steady fundamentals encourage holding, but when corrections occur, they are sharp — possibly triggered by large holders unwinding positions in liquid markets. The same pattern appears for TVL/market cap (rho = −0.14) and the tokenomics gauge (rho = −0.10).

For risk management, this implies that fundamentally strong protocols are not low-risk in the crash-magnitude sense. A trader using stop-losses should complement fundamental scores with derivatives-based metrics (open interest, funding rate) for position management.

4.5 Funding Rate and Bug Bounty: Niche but Stable Skew Predictors

Two unexpected scores emerge as stable skew predictors. Funding rate (rho = +0.095, sign = 100%): positive funding rates — indicating more long than short demand in perpetual futures — predict positive skew. When the market is net-long, occasional short squeezes produce outsized upward moves.

Bug bounty presence (rho = +0.086, sign = 100%, std = 0.013): projects offering bug bounties exhibit positive skew with the lowest standard deviation of any signal in this study. The mechanism may relate to selection effects: projects that invest in security tend to be more established, with price behavior driven by fundamental adoption events rather than exploit-driven crashes.

4.6 Trending Rank: The Hype Warning Signal

Trending rank shows consistent negative correlation with stop-loss survival (rho = −0.09 to −0.13, sign = 100%). Cryptocurrencies that are currently trending are more likely to trigger stop-losses before reaching take-profit targets. This aligns with the well-documented "buy the hype, sell the news" pattern: trending assets attract late buyers near local peaks, and the subsequent mean-reversion hits stop-losses before any new upside materializes.