Paper: arXiv 2501.12841
Authors: Heming Chen, Xiaojing Cai
Abstract
This study investigates three central questions in portfolio optimization. First, whether time-varying moment estimators outperform conventional sample estimators in practical portfolio construction. Second, whether incorporating a turnover penalty into the optimization objective can improve out-of-sample performance. Third, what type of optimal portfolio strategies can consistently outperform the naive 1/N benchmark. Using empirical evidence from the cryptocurrencies market, this paper provides comprehensive answers to these questions. In the process, several additional findings are uncovered, offering further insights into the dynamics of portfolio construction in highly volatile asset classes.
Complexity vs Empirical Score
- Math Complexity: 6.5/10
- Empirical Rigor: 7.5/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper employs advanced statistical models like DCC-EGARCH and deep-learning forecasting, indicating high mathematical complexity. It also demonstrates strong empirical rigor through extensive backtesting on cryptocurrency data, considering transaction costs and multiple frequencies.
Research Flowchart
flowchart TD
A["Research Goal: Determine optimal portfolio strategies<br>in crypto markets with time-varying moments & costs"] --> B["Data: Crypto Asset Returns<br>Historical Price Data"]
B --> C{"Key Methodology Tests"}
C --> D["Estimators Comparison<br>Time-Varying vs. Sample Moments"]
C --> E["Objective Optimization<br>Standard vs. Turnover Penalty"]
C --> F["Benchmarking<br>Naive 1/N vs. Optimal Strategies"]
D & E & F --> G["Computational Process<br>Rolling Window Backtesting<br>Transaction Cost Modeling"]
G --> H["Key Findings"]
subgraph H ["Outcomes"]
H1["Time-varying moments generally outperform sample estimators"]
H2["Turnover penalty improves risk-adjusted returns by reducing trading frequency"]
H3["Optimal strategies with low-frequency rebalancing beat naive 1/N consistently"]
end
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 compares four optimization frameworks (MV, GMV, MVC, GMVC) and evaluates time‑varying moment estimators and turnover penalties in cryptocurrency portfolios.
Source sentence
By incorporating both analytical derivations and empirical evidence under the mean–variance analysis, we compare four optimization frameworks (MV, GMV, MVC, and GMVC), and employ different rebalancing frequencies and risk‑aversion coefficients to ensure the robustness of our results.
Assumes 6
- Excess returns follow a multivariate normal distribution. model
Source sentence
Under the assumption that excess return follows a multivariate normal distribution, by using an expected loss function in the standard mean–variance analysis context, Kan and Zhou [2007] analytically show...
- Short‑selling is prohibited (weights non‑negative). execution
Source sentence
we impose the following short selling constraint across all four optimization frameworks mentioned above in the empirical analysis presented below: wi ≥0, ∀i
- The first 70% of the data is used for training and the remaining 30% for out‑of‑sample testing. data
Source sentence
Specifically, regardless of daily or weekly returns, the first 70% of the dataset is used as the training set to fit the multivariate GARCH model and the remaining 30% is used as the testing set for out-of-sample forecasting and evaluation.
- Proportional transaction costs are 50 basis points per asset. execution
Source sentence
where β = 0.005 represents proportional transaction costs of 50 basis points for each of the risky assets [DeMiguel et al., 2009]
Show 2 more
- Risk‑aversion coefficients are examined in the range 1 to 10. model
Source sentence
we evaluate portfolio performance using risk‑aversion coefficients ranging from 1 to 10 for all optimal strategies
- Returns are modeled with a multivariate Student’s t distribution within a GARCH framework. statistical
Source sentence
it is appropriate to model the volatility dynamics using the multivariate GARCH framework described in Section 2, assuming a multivariate Student’s t distribution.
Data
- Cryptocurrency price dataset market-prices public
Source sentence
The dataset employed in this study comprises daily and weekly returns for the four longstanding and most liquid cryptocurrencies over the entire observation period: Bitcoin, Ethereum, Ripple and Litecoin, obtained from http://www.coinmarketcap.com.
Costs
Modeled
Proportional transaction costs of 0.005 (50 bps) per asset are included in the MVC optimization objective and reported in out‑of‑sample results.
Source sentence
where β = 0.005 represents proportional transaction costs of 50 basis points for each of the risky assets [DeMiguel et al., 2009]
Finds 5
- Time‑varying estimators of means and covariances are inferior to sample‑based estimators, leading to higher turnover and transaction costs without improving allocation. performance
Source sentence
Our empirical findings reveal that time‑varying estimators of the first and second moments, such as deep‑learning forecasts and DCC covariances, are inferior to their sample‑based counterparts.
- The turnover penalty reduces transaction costs across all strategies but does not consistently improve overall performance. performance
Source sentence
our results show that it effectively reduces transaction costs across all optimal strategies but does not necessarily enhance the final performance of every strategy.
- Most volatility‑timing strategies earn positive performance fees and outperform the naive 1/N benchmark. performance
Source sentence
Even in the highly correlated cryptocurrencies market, most volatility‑timing strategies earn positive performance fees and outperform the naive 1/N benchmark
- Strategies based on mean forecasts underperform both volatility‑timing strategies and the naive 1/N benchmark. negative
Source sentence
portfolio strategies depending on mean forecasts underperform both the volatility‑timing strategies and the naive 1/N benchmark
- The benefit of the turnover penalty diminishes as the rebalancing frequency decreases. theoretical
Source sentence
the impact of imposing a turnover penalty diminishes as the rebalancing frequency decreases, consistent with the theoretical deduction of Woodside‑Oriakhi et al. [2013]
Checked by others
Supported by
- Black-Litterman and ESG Portfolio Optimization
Compare the findings
This paper: The turnover penalty reduces transaction costs across all strategies but does not consistently improve overall performance.
That paper: BL portfolios with the same turnover constraint often incurred excessive turnover, making them infeasible, whereas increasing the turnover penalty reduced turnover but also lowered returns.
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 (51)
- Optimal versus naive diversification: How inefficient is the 1/N portfolio strategy? (2009)
- Optimal vs na "\i ve diversification in cryptocurrencies (2018)
- Markowitz meets Talmud: A combination of sophisticated and naive diversification strategies (2011)
- It’s all in the timing: simple active portfolio strategies that outperform naive diversification (2012)
- Optimal portfolio choice with parameter uncertainty (2007)
- 60 years of portfolio optimization: Practical challenges and current trends (2014)
- An economic evaluation of empirical exchange rate models (2009)
- Optimal and naive diversification in currency markets (2017)
and 43 more
- Diversification in financial and crypto markets (2023)
- Statistical, machine learning and deep learning forecasting methods: Comparisons and ways forward (2023)
- The mean-variance approach to portfolio optimization subject to transaction costs (1996)
- The Effect of Errors in Means, Variances, and Covariances on Optimal Portfolio Choice (1993)
- Gluonts: Probabilistic time series models in python (2019) — link
- DeepAR: Probabilistic forecasting with autoregressive recurrent networks (2020)
- Asymmetric volatility in cryptocurrency markets: New evidence from smooth transition GARCH models (2020)
- High frequency volatility co-movements in cryptocurrency markets (2019)
- Dynamic conditional correlation: A simple class of multivariate generalized autoregressive conditional heteroskedasticity models (2002)
- Conditional heteroskedasticity in asset returns: A new approach (1991)
- Model-based pairs trading in the bitcoin markets (2017)
- The Markowitz optimization enigma: Is ‘optimized’optimal? (1989)
- Do commodities add economic value in asset allocation? New evidence from time-varying moments (2018)
- Correlation dynamics and international diversification benefits (2014)
- Algorithms for hyper-parameter optimization (2011)
- Does model complexity add value to asset allocation? Evidence from machine learning forecasting models (2022)
- A robust perspective on transaction costs in portfolio optimization (2018)
- Large-scale portfolio allocation under transaction costs and model uncertainty (2019)
- The economic value of predicting stock index returns and volatility (2004)
- Asset allocation with a high dimensional latent factor stochastic volatility model (2006)
- A utility-based comparison of some models of exchange rate volatility (1993)
- The economic value of volatility timing (2001)
- When “time varying” volatility meets “transaction cost” in portfolio selection (2023)
- Cryptocurrency price prediction using traditional statistical and machine-learning techniques: A survey (2021)
- Forecasting the price of Bitcoin using deep learning (2021)
- Performance evaluation of deep learning and boosted trees for cryptocurrency closing price prediction (2023)
- Cryptocurrency price forecasting–A comparative analysis of ensemble learning and deep learning methods (2024)
- Deep learning and technical analysis in cryptocurrency market (2023)
- Probabilistic deep learning and transfer learning for robust cryptocurrency price prediction (2024)
- Artificial Intelligence Alter Egos: Who might benefit from robo-investing? (2020)
- Mean–variance portfolio optimization using machine learning-based stock price prediction (2021)
- Portfolio optimization with return prediction using deep learning and machine learning (2021)
- Mean–variance portfolio optimization with deep learning based-forecasts for cointegrated stocks (2022)
- DeepAR-Attention probabilistic prediction for stock price series (2024)
- Criteria for classifying forecasting methods (2020)
- Global currency hedging with common risk factors (2020)
- Can currency-based risk factors help forecast exchange rates? (2016)
- The economic value of volatility timing using “realized” volatility (2003)
- Portfolio rebalancing with an investment horizon and transaction costs (2013)
- Where do we stand in cryptocurrencies economic research? A survey based on hybrid analysis (2021)
- Risks and returns of cryptocurrency (2021)
- The predictive power of public Twitter sentiment for forecasting cryptocurrency prices (2020)
- Directional predictability between returns and volume in cryptocurrencies markets (2021)
0 of 51 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.