Paper: arXiv 2512.02029

Authors: Weikang Zhang, Alison Watts

Abstract

Crypto enthusiasts claim that buying and holding crypto assets yields high returns, often citing Bitcoin’s past performance to promote other tokens and fuel fear of missing out. However, understanding the real risk-return trade-off and what factors affect future crypto returns is crucial as crypto becomes increasingly accessible to retail investors through major brokerages. We examine the HODL strategy through two independent analyses. First, we implement 480 million Monte Carlo simulations across 378 non-stablecoin crypto assets, net of trading fees and the opportunity cost of 1-month Treasury bills, and find strong evidence of survivorship bias and extreme downside concentration. At the 2-3 year horizon, the median excess return is -28.4 percent, the 1 percent conditional value at risk indicates that tail scenarios wipe out principal after all costs, and only the top quartile achieves very large gains, with a mean excess return of 1,326.7 percent. These results challenge the HODL narrative: across a broad set of assets, simple buy-and-hold loads extreme downside risk onto most investors, and the miracles mostly belong to the luckiest quarter. Second, using a Bayesian multi-horizon local projection framework, we find that endogenous predictors based on realized risk-return metrics have economically negligible and unstable effects, while macro-finance factors, especially the 24-week exponential moving average of the Fear and Greed Index, display persistent long-horizon impacts and high cross-basket stability. Where significant, a one-standard-deviation sentiment shock reduces forward top-quartile mean excess returns by 15-22 percentage points and median returns by 6-10 percentage points over 1-3 year horizons, suggesting that macro-sentiment conditions, rather than realized return histories, are the dominant indicators for future outcomes.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 9.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced mathematical methods such as 480 million Monte Carlo simulations, Bayesian multi-horizon local projection, and Bayesian hierarchical random-effects meta-analysis, demonstrating high mathematical density. It also shows strong empirical rigor through massive-scale backtesting across 378 assets, incorporation of real-world costs (fees, opportunity costs), and use of high-frequency macro-sentiment data.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Evaluate the HODL Strategy &<br>Identify Drivers of Crypto Returns"] --> B["Methodology Part 1:<br>480M Monte Carlo Simulations"]
  A --> C["Methodology Part 2:<br>Bayesian Multi-Horizon Local Projection"]
  
  B --> D{"Data: 378 Non-Stablecoin Assets<br>Inputs: Net of Fees & T-Bill Cost"}
  D --> E["Computational Process:<br>Simulate Buy & Hold Across Horizons"]
  E --> F["Outcome Part 1:<br>Survivorship Bias & Extreme Downside<br>Median Return: -28.4%<br>Top Quartile: +1,326.7%"]
  
  C --> G{"Data: Macro-Finance Factors<br>(e.g., Fear & Greed Index)"}
  G --> H["Computational Process:<br>Analyze Sentiment Impact on Future Returns"]
  H --> I["Outcome Part 2:<br>Macro-Sentiment Dominates History<br>1 Std. Dev. Shock ↓ Returns by 15-22%"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper conducts 480 million Monte‑Carlo simulations of buy‑and‑hold crypto returns and a Bayesian multi‑horizon local projection analysis of macro‑sentiment predictors.

Source sentence
First, we implement 480 million Monte Carlo simulations across 378 non-stablecoin crypto assets, net of trading fees and the opportunity cost of 1-month Treasury bills, and find strong evidence of survivorship bias

Assumes 5

  • Holding horizons are drawn uniformly from a discrete range of days. model
    Source sentence
    Given inclusive horizon bounds ℓ< h, τi ∼DiscreteUniform{ℓ, ℓ+ 1, . . . , h} (i = 1, . . . , n), where τi denotes the randomly drawn holding horizon for episode i.
  • Buy dates are uniformly sampled within the data window ensuring enough days for the horizon. execution
    Source sentence
    si ∼DiscreteUniform{0, 1, . . . , T −τi −1}, ei = si + τi.
  • Coins are selected uniformly from the set of 378 tradable tokens. data
    Source sentence
    ci ∼DiscreteUniform{0, . . . , C −1}, where ci denotes the coin selected for episode i.
  • Risk‑free benchmark is the continuously rolled 1‑month U.S. Treasury bill return over the same calendar days. model
    Source sentence
    we benchmark it against the return an investor would have earned by continuously rolling over the 1‑month U.S. Treasury bill across the exact same calendar days.
Show 1 more
  • Macro‑finance predictors are treated as exogenous and standardized for one‑standard‑deviation shocks. model
    Source sentence
    βh,p,y measures how target y changes in standardized units at horizon h when predictor p increases by one standard deviation at time t

Data

  • 378 tradable crypto assets (ALL basket) market-prices unknown
    378 non‑stablecoin crypto tokens
    Source sentence
    the ALL basket contains the entire dataset comprising C = 378 tradable tokens.
  • 1‑month U.S. Treasury bill daily returns benchmark unknown
    Source sentence
    we benchmark it against the return an investor would have earned by continuously rolling over the 1‑month U.S. Treasury bill across the exact same calendar days.
  • Crypto Fear & Greed Index (24‑week EMA) macro unknown
    Source sentence
    macro‑finance factors, especially the 24‑week exponential moving average of the Fear and Greed Index, display persistent long‑horizon impacts
  • Various macro series (Fed funds, Treasury yields, VIX, etc.) macro unknown
    Source sentence
    Table 5: Unit‑root tests for macro series…

Recognised sources: FRED

Costs

Modeled Costs are part of the reported headline results.

Finds 5

  • Median excess return over 2‑3 year horizons is –28.4 %, while the top‑quartile mean excess return reaches +1,326.7 %. performance
    Source sentence
    At the longest holding horizon (731–1095 days), the median excess return is −28.4%, yet the top‑quartile mean reaches +1326.7%.
  • The 1 % CVaR in tail scenarios wipes out the principal after all costs. other
    Source sentence
    the 1 percent conditional value at risk indicates that tail scenarios wipe out principal after all costs
  • A one‑standard‑deviation increase in the sentiment shock reduces forward top‑quartile mean excess returns by 15‑22 percentage points and median returns by 6‑10 percentage points over 1‑3 year horizons. predictive-accuracy
    Source sentence
    a one‑standard‑deviation sentiment shock reduces forward top‑quartile mean excess returns by 15-22 percentage points and median returns by 6-10 percentage points over 1-3 year horizons
  • For holding periods beyond 181‑365 days, the probability of losses greater than 10 % exceeds 50 % and CVaR 1 % approaches near‑total loss. other
    Source sentence
    for holding periods beyond 181–365 days, the probability of experiencing losses greater than 10% exceeds 50%, and the CVaR1% approaches near‑total loss (100%).
  • Realized risk‑return metrics have economically negligible and unstable effects, whereas macro‑sentiment predictors show persistent impacts across horizons. statistical
    Source sentence
    endogenous predictors based on realized risk‑return metrics have economically negligible and unstable effects, while macro‑finance factors

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 30 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (44)

  1. Efficient tests for an autoregressive unit root (1996) — link
  2. Testing the null hypothesis of stationarity against the alternative of a unit root: How sure are we that economic time series have a unit root? (1992) — link
  3. Further evidence on the great crash, the oil-price shock, and the unit-root hypothesis (1992) — link
  4. An introduction to long-memory time series models and fractional differencing (1980) — link
  5. Fractional differencing (1981) — link
  6. An application of extreme value theory to cryptocurrencies (2018) — link
  7. A statistical risk assessment of Bitcoin and its extreme tail behavior (2017) — link
  8. Bitcoin versus S&P 500 Index: Return and Risk Analysis — → 2310.02436
and 36 more
  1. The accuracy of risk measurement models on Bitcoin market during COVID-19 pandemic (2021) — link
  2. Does size matter in the cryptocurrency market? (2020) — link
  3. Enduring relief or fleeting respite? Bitcoin as a hedge and safe haven for the US dollar (2024) — link
  4. MAX momentum in cryptocurrency markets (2021) — link
  5. Stocks for the long run? Evidence from a broad sample of developed markets (2022) — link
  6. Time–frequency quantile dependence between Bitcoin and global equity markets (2021) — link
  7. Estimation and Inference of Impulse Responses by Local Projections (2005) — link
  8. How reliable are local projection estimators of impulse responses? (2011) — link
  9. Bayesian local projections (2025) — link
  10. Impulse response estimation by smooth local projections (2019) — link
  11. Heuristics of instability and stabilization in model selection (1996) — link
  12. Variable selection with error control: Another look at stability selection (2013) — link
  13. Stability selection (2010) — link
  14. Multitask group Lasso for Genome Wide Association Studies in diverse populations (2022) — link
  15. Sparse multitask group Lasso for genome-wide association studies (2025) — link
  16. The horseshoe estimator for sparse signals (2010) — link
  17. Sparsity information and regularization in the horseshoe and other shrinkage priors (2017) — link
  18. On the Hyperprior Choice for the Global Shrinkage Parameter in the Horseshoe Prior (2017)
  19. Monetary policy shocks and Bitcoin prices (2022) — link
  20. Monetary policy and Bitcoin (2023) — link
  21. The Impact of Investor Sentiment on Bitcoin Returns and Conditional Volatilities during the Era of Covid-19 (2023) — link
  22. The Stationary Bootstrap (1994) — link
  23. Advances in Financial Machine Learning (2018)
  24. Regularization and Variable Selection via the Elastic Net (2005) — link
  25. Scikit-learn: Machine Learning in P ython (2011)
  26. Resampling Methods for Dependent Data (2003) — link
  27. On blocking rules for the bootstrap with dependent data (1995) — link
  28. Bayesian Inference of Local Projections with Roughness Penalty Priors (2020) — link
  29. General seemingly unrelated local projections (2024) — link
  30. Bayesian Treatment of the Independent Student-t Linear Model (1993)
  31. A Heteroskedasticity-Consistent Covariance Matrix Estimator and a Direct Test for Heteroskedasticity (1980) — link
  32. Some heteroskedasticity-consistent covariance matrix estimators with improved finite sample properties (1985)
  33. Control of generalized error rates in multiple testing (2007) — link
  34. Bayesian methods in meta-analysis and evidence synthesis (2001)
  35. Bayesian Random-Effects Meta-Analysis Using the bayesmeta R Package (2020) — link
  36. The No-U-Turn Sampler: Adaptively Setting Path Lengths in Hamiltonian Monte Carlo (2014)

1 of 44 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.