Paper: arXiv 2312.03294
Authors: Tuoyuan Cheng, Kan Chen
Abstract
In this paper, we present an integrated approach to portfolio construction and optimization, leveraging high-performance computing capabilities. We first explore diverse pairings of generative model forecasts and objective functions used for portfolio optimization, which are evaluated using performance-attribution models based on LASSO. We illustrate our approach using extensive simulations of crypto-currency portfolios, and we show that the portfolios constructed using the vine-copula generative model and the Sharpe-ratio objective function consistently outperform. To accommodate a wide array of investment strategies, we further investigate portfolio blending and propose a general framework for evaluating and combining investment strategies. We employ an extension of the multi-armed bandit framework and use value models and policy models to construct eclectic blended portfolios based on past performance. We consider similarity and optimality measures for value models and employ probability-matching (“blending”) and a greedy algorithm (“switching”) for policy models. The eclectic portfolios are also evaluated using LASSO models. We show that the value model utilizing cosine similarity and logit optimality consistently delivers robust superior performances. The extent of outperformance by eclectic portfolios over their benchmarks significantly surpasses that achieved by individual generative model-based portfolios over their respective benchmarks.
Complexity vs Empirical Score
- Math Complexity: 8.0/10
- Empirical Rigor: 7.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper utilizes advanced statistical and machine learning methods like vine copulas, multi-armed bandits, and LASSO, reflecting high mathematical complexity. It is also data-heavy, employing extensive simulations on cryptocurrency data and real-world performance evaluation, indicating substantial empirical rigor.
Research Flowchart
flowchart TD A["Research Goal:<br>Develop General Framework for<br>Portfolio Construction & Optimization"] --> B["Methodology: Simulation & Blending"] B --> C["Input: Cryptocurrency Price Data"] C --> D["Computational Process 1:<br>Generative Model Forecasts<br>Vine-Copula vs. Others"] D --> E["Computational Process 2:<br>Portfolio Optimization<br>Sharpe-Ratio Objective"] E --> F["Computational Process 3:<br>Portfolio Blending<br>Multi-Armed Bandit Framework"] F --> G["Performance Evaluation:<br>LASSO Performance Attribution"] G --> H["Key Finding:<br>Vine-Copula + Sharpe Ratio<br>outperforms benchmarks"] G --> I["Key Finding:<br>Eclectic Blended Portfolios<br>deliver robust superior performance"]
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 an integrated, high‑performance computing framework for constructing and optimizing cryptocurrency portfolios using generative‑model forecasts, objective functions, and a multi‑armed bandit‑based blending of strategies.
Source sentence
We present an integrated approach to portfolio construction and optimization, leveraging high-performance computing capabilities.
Assumes 6
- Rebalancing is performed every two days. execution
Source sentence
Rebalancing occurs every two days, targeting diverse combinations of generative dependence model forecasts and proxy objective functions.
- A flat transaction cost of 50 basis points per trade is applied. execution
Source sentence
The transaction cost c is set as 50 bps, together with the number of assets D = 12 and the constraint boundary multiplier m = 5.
- A rolling window of 91 steps (182 days) is used to fit generative models. model
Source sentence
To fit the generative model, we employ a rolling window length of 91 steps or 182 days.
- Historical price series from Binance and Coinbase are representative of cryptocurrency markets. data
Source sentence
We retrieved price time series from Binance, spanning from 2018-06-10 to 2023-06-04, covering 12 currencies
Show 2 more
- The Kelly‑type objective function assumes a constant transaction cost aversion coefficient v. model
Source sentence
c denotes the constant transaction, and v is a transaction cost aversion coefficient, which is used to control portfolio turnover.
- Asset return dependence can be captured by vine‑copula or other generative models fitted by maximum likelihood. model
Source sentence
The regular vine copula structure is constructed by picking the maximum spanning tree, which maximizes the sum of absolute values of Kendall’s τ among pairwise variables while adhering to proximity conditions.
Data
- Binance cryptocurrency price series market-prices public
Source sentence
We retrieved price time series from Binance, spanning from 2018-06-10 to 2023-06-04, covering 12 currencies priced in USDT including ADA, BNB, BTC, EOS, ETH, IOTA, LTC, NEO, ONT, QTUM, XLM, and XRP.
- Coinbase cryptocurrency price series market-prices public
Source sentence
spanning from 2019-08-08 to 2023-08-17, covering 10 currencies priced in USD including BCH, BTC, EOS, ETC, ETH, LINK, LTC, XLM, XTZ, and ZRX.
Costs
Modeled
A flat cost of 50 basis points per trade is incorporated in all backtests.
Source sentence
The transaction cost c is set as 50 bps, together with the number of assets D = 12 and the constraint boundary multiplier m = 5.
Finds 5
- Portfolios built with the vine‑copula generative model and Sharpe‑ratio objective consistently outperformed benchmarks and other model‑objective pairings. performance
Source sentence
The Sharpe ratio portfolios, as classical finite‑moments portfolios, outperformed Kelly’s portfolios, quantile portfolios, and parity portfolios.
- The vine copula emerged as the superior choice for multivariate dependence modeling across all experiments. performance
Source sentence
Notably, the vine copula emerged as a superior choice for multivariate dependence modeling, and Sharpe ratio portfolios consistently outperformed alternative approaches.
- Eclectic portfolios using cosine similarity and logit optimality delivered robust superior performance, surpassing individual generative‑model portfolios. performance
Source sentence
We show that the value model utilizing cosine similarity and logit optimality consistently delivers robust superior performances.
- Switching (greedy) eclectic portfolios exhibited higher variability but achieved higher average performance than blending (probability‑matching) portfolios. performance
Source sentence
Switching portfolios exhibited higher variability but also achieved superior average performance compared to blending portfolios.
- A moderate transaction‑cost aversion coefficient (v around 2–3) improves terminal profits, whereas too high a cost barrier limits returns. performance
Source sentence
A moderate transaction cost aversion coefficient v around 2 to 3 is favorable, but too high a hurdle for trading could limit terminal profits.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 18 related papers). Coverage is partial, so this is neither replication nor refutation.
2 papers in the archive : MILLION: A General Multi-Objective Framework with …
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 (62)
- An experimental application of the Delphi method to the use of experts (1963)
- Multiple Portfolio Blending Strategy with Thompson Sampling (2022)
- Measuring tail risks (2022) — link
- Spectral measures of risk: A coherent representation of subjective risk aversion (2002) — link
- Information theory and an extension of the maximum likelihood principle (1998) — link
- Gain, loss, and asset pricing (2000) — link
- Generalized autoregressive conditional heteroskedasticity (1986) — link
- Analyzing dependent data with vine copulas: A practical guide with R (2019) — link
and 54 more
- Vine copula based dependence modeling in sustainable finance (2022) — link
- The 10 reasons most machine learning funds fail (2018) — link
- Selecting and estimating regular vine copulae and application to financial returns (2013) — link
- Least angle regression (2004) — link
- Dynamic conditional correlation: A simple class of multivariate generalized autoregressive conditional heteroskedasticity models (2002) — link
- On the diversity of equity markets (1999) — link
- Stochastic portfolio theory: An overview (2009) — link
- The utility analysis of choices involving risk (1948) — link
- Quantifying and using expert opinion for variable-selection problems in regression (1996) — link
- Decision-theoretic foundations for causal reasoning (1995)
- The decision to seed hurricanes (1972) — link
- COBra: Copula-Based Portfolio Optimization (2018)
- Fallibility, reflexivity, and the human uncertainty principle (2013)
- In pursuit of the perfect portfolio (2021) — link
- On the Likelihood that One Unknown Probability Exceeds Another in View of the Evidence of Two Samples (1933)
- Reinforcement learning: An introduction (2018)
- Remarks on a multivariate transformation (1952)
- The pillars of science (2022)
- Optimization of conditional value-at-risk (2000)
- Riskmetrics—technical document (1996)
- Mutual Fund Performance (1966)
- Performance measurement in a downside risk framework (1994) — link
- Portfolio Choice and the Kelly Criterion (1975) — link
- A New Interpretation of Information Rate (1956) — link
- Scikit-Learn: Machine Learning in Python (2011)
- Quantitative Portfolio Management: The Art and Science of Statistical Arbitrage (2021)
- Portfolio Blending via Thompson Sampling (2016)
- To maximize or randomize? An experimental study of probability matching in financial decision making (2021) — link
- Mean–variance approximations to expected utility (2014) — link
- 60 Years of portfolio optimization: Practical challenges and current trends (2014) — link
- Statistical rethinking: A Bayesian course with examples in R and Stan (2020)
- Portfolio selection (1952) — link
- Safety first and the holding of assets (1952)
- The Early History of Portfolio Theory: 1600-1960 (1999)
- Markowitz’s “portfolio selection”: A fifty-year retrospective (2002)
- de Finetti scoops Markowitz (2006)
- Portfolio theory: As I still see it (2010)
- The foundations of mathematical and other logical essays (1931)
- Theory of Games and Economic Behavior (60th Anniversary Commemorative Edition) (1944)
- The Theory of Statistical Decision (1951) — link
- Causal decision theory (1981) — link
- Causal diagrams for empirical research (1995) — link
- Alternatives to Expected Utility: Formal Theories (2004) — link
- Black–Litterman and beyond: The bayesian paradigm in investment management (2021)
- On the financial interpretation of risk contribution: Risk budgets do add up (2005)
- Higher-moment portfolio theory (2005) — link
- Robust portfolio optimization: a categorized bibliographic review (2020) — link
- Poor Charlie ’s Almanack (2008)
- Multi-period portfolio optimization (2022)
- DCC - GARCH model-with various error distributions (2009)
- Dependence Modeling with Copulas (2014)
- Risk parity and diversification (2011)
- Adaptive markets: Financial evolution at the speed of thought (2017) — link
- Prospect Theory : An Analysis of Decision under Risk (1979) — link
0 of 62 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.