Paper: arXiv 2508.15922
Authors: Grzegorz Dudek, Witold Orzeszko, Piotr Fiszeder
Abstract
Cryptocurrency markets are characterized by extreme volatility, making accurate forecasts essential for effective risk management and informed trading strategies. Traditional deterministic (point) forecasting methods are inadequate for capturing the full spectrum of potential volatility outcomes, underscoring the importance of probabilistic approaches. To address this limitation, this paper introduces probabilistic forecasting methods that leverage point forecasts from a wide range of base models, including statistical (HAR, GARCH, ARFIMA) and machine learning (e.g. LASSO, SVR, MLP, Random Forest, LSTM) algorithms, to estimate conditional quantiles of cryptocurrency realized variance. To the best of our knowledge, this is the first study in the literature to propose and systematically evaluate probabilistic forecasts of variance in cryptocurrency markets based on predictions derived from multiple base models. Our empirical results for Bitcoin demonstrate that the Quantile Estimation through Residual Simulation (QRS) method, particularly when applied to linear base models operating on log-transformed realized volatility data, consistently outperforms more sophisticated alternatives. Additionally, we highlight the robustness of the probabilistic stacking framework, providing comprehensive insights into uncertainty and risk inherent in cryptocurrency volatility forecasting. This research fills a significant gap in the literature, contributing practical probabilistic forecasting methodologies tailored specifically to cryptocurrency markets.
Complexity vs Empirical Score
- Math Complexity: 6.5/10
- Empirical Rigor: 8.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced statistical and machine learning methods, including quantile regression forests and multi-model stacking frameworks, requiring dense mathematical formulation. It demonstrates high empirical rigor with detailed data processing (log-transformed realized variance), a comprehensive backtest using Bitcoin data from 2017-2021, and robust evaluation metrics (CRPS, Winkler Score).
Research Flowchart
flowchart TD
A["Research Goal<br>Forecast Cryptocurrency Volatility<br>Using Probabilistic Methods"] --> B["Data Input<br>Bitcoin Realized Variance<br>(Log-Transformed)"]
B --> C{"Methodology"}
C --> D["Base Model Processing<br>Statistical: HAR, GARCH, ARFIMA<br>ML: LASSO, SVR, MLP, RF, LSTM"]
D --> E["Probabilistic Framework<br>Quantile Estimation via Residual Simulation<br>- Linear Models<br>- Stacking Ensemble"]
E --> F["Computational Process<br>Simulate Residuals<br>Generate Quantile Predictions<br>Estimate Conditional Distribution"]
F --> G["Key Findings & Outcomes"]
G --> H["QRS Method with Linear Base Models<br>Optimal for Log-Volatility Data"]
G --> I["Robust Probabilistic Stacking<br>Comprehensive Risk Assessment"]
G --> J["Significant Literature Gap Filled<br>Practical Methodology for Crypto Markets"]
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 proposes probabilistic forecasting methods that convert point forecasts from multiple base models into quantile‑based forecasts, notably the Quantile Estimation through Residual Simulation (QRS) approach.
Source sentence
We propose approaches that transform deterministic point forecasts of cryptocurrency volatility into probabilistic (quantile‑based) forecasts, capturing the inherent uncertainty in cryptocurrency markets.
Assumes 3
- Past residuals accurately represent future forecast uncertainty. statistical
Source sentence
It assumes that past residuals accurately represent future uncertainty, which may not hold under volatile conditions.
- Log‑transformation stabilizes variance and makes data more amenable to linear modeling. data
Source sentence
the log transformation stabilizes variance and makes the data more amenable to linear modeling
- The residual distribution is stationary over the forecast horizon. statistical
Source sentence
In our implementation, we attempt to mitigate this limitation by extending the set Λ up to the forecasted day τ.
Data
- Bitcoin (BTC/USD) data market-prices public
Source sentence
The dataset was obtained from the crypto exchange Kraken and covers the period from 2017 to 2021.
Kraken
Costs
Not applicable
Finds 4
- The QRS‑l models (e.g., HAR‑l, HAR‑R‑l, ARFIMA‑l, etc.) are statistically more accurate in terms of CRPS than other models, making them the best‑performing approaches. performance
Source sentence
Notably, the best‑performing models in this comparison belong to the QRS‑l category, including HAR‑l, HAR‑R‑l, AFRIMA‑l, RR‑l, SVR‑G‑l, SVR‑L‑l, and Ens‑Med‑l.
- QRS applied to linear base models on log‑transformed realized volatility consistently outperforms more sophisticated alternatives. predictive-accuracy
Source sentence
Our empirical results for Bitcoin demonstrate that the Quantile Estimation through Residual Simulation (QRS) method, particularly when applied to linear base models operating on log‑transformed realized volatility data, consistently outperforms more sophisticated alternatives.
- Probabilistic approaches do not substantially improve point‑forecast MAE/MSE compared to deterministic base models, with only slight improvements for a few models. negative
Source sentence
Comparing MAE‑Q and MSE‑Q (Table I) with the MAE and MSE values obtained from the point forecasts of base models
- QRS is the fastest method, requiring only 0.015 s per run, whereas QLR and QRF are slower. performance
Source sentence
tiles) on the machine specified above, we obtained approximate execution times of: 0.015 s for QRS, 0.8 s for QLR, and 0.4 s for QRF.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 39 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.