Paper: arXiv 2410.23297
Authors: Hugo Inzirillo
Abstract
We propose a new way of building portfolios of cryptocurrencies that provide good diversification properties to investors. First, we seek to filter these digital assets by creating some clusters based on their path signature. The goal is to identify similar patterns in the behavior of these highly volatile assets. Once such clusters have been built, we propose “optimal” portfolios by comparing the performances of such portfolios to a universe of unfiltered digital assets. Our intuition is that clustering based on path signatures will make it easier to capture the main trends and features of a group of cryptocurrencies, and allow parsimonious portfolios that reduce excessive transaction fees. Empirically, our assumptions seem to be satisfied.
Complexity vs Empirical Score
- Math Complexity: 8.5/10
- Empirical Rigor: 6.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper introduces mathematically advanced path signature theory (rough paths, iterated integrals) to finance, scoring high on math complexity. It includes a clear empirical setup with portfolio construction and comparative backtests on cryptocurrencies, though it lacks public code/datasets or heavy statistical validation, placing it above moderate empirical rigor.
Research Flowchart
flowchart TD RQ["Research Goal:<br>Develop crypto portfolio construction<br>via clustering for diversification"] --> Input["Data Input:<br>Historical price data<br>of digital assets"] Input --> PathSig["Methodology:<br>Compute Path Signatures<br>for each asset time series"] PathSig --> Cluster["Methodology:<br>Apply Clustering algorithm<br>on Path Signatures"] Cluster --> Portfolios["Computation:<br>Construct portfolios<br>from clusters"] Portfolios --> Eval["Computation:<br>Compare performance to<br>unfiltered universe"] Eval --> Outcomes["Key Findings:<br>Clustering enables parsimonious,<br>trend-capturing portfolios with<br>good diversification and lower fees"]
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 building cryptocurrency portfolios by clustering digital assets using path signatures to achieve diversification and lower transaction costs.
Source sentence
We propose a new way of building portfolios of cryptocurrencies that provide good diversification properties to investors. First, we seek to filter these digital assets by creating some clusters based on their path signature.
Assumes 5
- Liquidity of selected digital assets is sufficient to replicate the strategy in the real world. market
Source sentence
Thus, we are confident that the liquidity of each digital asset is sufficient to replicate the strategy in the real world.
- Recent past price behaviors offer the best indications of future trends. model
Source sentence
By using a sliding window, we assume that recent past price behaviors offer the best indications of future trends.
- Rebalancing can be executed at closing times with no slippage. execution
Source sentence
For each portfolio, we assume we are able to execute rebalancing at closing times (00:00 UTC).
- Trading fees are 20 basis points per trade. execution
Source sentence
For our backtest we used trading fees of 20 bp.
Show 1 more
- k-means clustering on path signatures yields meaningful asset groups. model
Source sentence
To perform this task, we rely on the k-means algorithm, that is applied to the path signatures of each of the selected digital assets.
Data
top 30 digital assets (cryptocurrencies) · crypto · weekly
- Binance daily closing price dataset market-prices public
Source sentence
In our experiment, we used a dataset downloaded from Binance. This dataset includes the top 30 digital assets in terms of market capitalisation, according to Coinmarketcap. We downloaded the daily closing price data for all the trading pairs "crypto against USD", and use USDT as a proxy of the US dollar.
Costs
Partial
A flat fee of 20 basis points per trade is assumed, but not incorporated into the headline performance results.
Source sentence
For our backtest we used trading fees of 20 bp. Considering that asset managers may play these strategies we did not considerered a fixed trading fees for each trading, primarily because of the unrealism it suggests.
Finds 3
Sharpe 1.68 (benchmark 0.73) · annual return 125.2% · max drawdown 61.6%
- The equally weighted portfolio using rolling‑window signature clustering (EW RW) achieved the highest annualised return (1.2523) and a high Sharpe (1.6849) and Calmar (2.0306) ratio, outperforming both the unfiltered EW and the fixed‑origin‑time clustered EW. performance
Source sentence
we observe a relatively high Sharpe and Calmar ratios (1.6849 and 2.0306, respectively).
- Signature‑based clustered strategies exhibit lower maximum drawdown than the unfiltered equally weighted portfolio. performance
Source sentence
Signature-based clustered stratgies exhibit lower Maximum DrawDown (MDD) than (EW)... suggesting that the former strategies have experienced smaller peak-to-trough declines during our studied period.
- The clustered portfolios require fewer trades, implying lower trading costs compared to the unfiltered strategy. performance
Source sentence
On both figures we can clearly identify the less expensive strategies in term of trading cost.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 12 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 (220)
- Hierarchical attention networks for document classification (2016)
- Dynamic detr: End-to-end object detection with dynamic attention (2021)
- Effective approaches to attention-based neural machine translation (2015) — link
- Google’s neural machine translation system: Bridging the gap between human and machine translation (2016) — link
- Sparse sinkhorn attention (2020)
- Efficient content-based sparse attention with routing transformers (2021)
- Heteroskedasticity in stock return data: Volume versus GARCH effects (1990)
- Regime changes in Bitcoin GARCH volatility dynamics (2019)
and 212 more
- Marginal likelihood for Markov-switching and change-point GARCH models (2014)
- GARCH modelling of cryptocurrencies (2017)
- Anomaly detection with conditional variational autoencoders (2019)
- Anomaly detection with robust deep autoencoders (2017)
- Anomaly detection using autoencoders with nonlinear dimensionality reduction (2014)
- Extreme financial risks: From dependence to risk management (2006)
- The choice of the distribution of asset returns: How extreme value theory can help? (2005)
- Why stock markets crash: critical events in complex financial systems (2009)
- Cluster analysis on the structure of the cryptocurrency market via Bitcoin–Ethereum filtering (2019)
- Bitcoin: A peer-to-peer electronic cash system (2008)
- Bitcoin-asset or currency? revealing users’ hidden intentions (2014)
- What are the main drivers of the Bitcoin price? Evidence from wavelet coherence analysis (2015)
- A new approach to linear filtering and prediction problems (1960)
- A simple way to initialize recurrent networks of rectified linear units (2015) — link
- Rectified linear units improve restricted boltzmann machines (2010)
- A cryptocurrency empirical study focused on evaluating their distribution functions (2022)
- Studying the properties of the Bitcoin as a diversifying and hedging asset through a copula analysis: Constant and time-varying (2020)
- Volatility estimation for Bitcoin: A comparison of GARCH models (2017)
- Accounting for Bitcoin and Other Cryptocurrencies under IFRS: A Comparison and Assessment of Competing Models (2018)
- Bitcoin: A Peer-to-Peer Electronic Cash System (2009)
- Can cryptocurrencies be a safe haven: a tail risk perspective analysis (2018)
- Bubble regime identification in an attention-based model for Bitcoin and Ethereum price dynamics (2020)
- The digital traces of bubbles: Feedback cycles between socio-economic signals in the Bitcoin economy (2014)
- Speculative bubbles in Bitcoin markets? An empirical investigation into the fundamental value of Bitcoin (2015)
- Deep State Space Models for Time Series Forecasting (2018)
- Attentive State-Space Modeling of Disease Progression (2019)
- Attention Is All You Need (2017)
- A New Approach to the Economic Analysis of Nonstationary Time Series and the Business Cycle (1989)
- Regime switching with time-varying transition probabilities (1993)
- Nonlinear principal component analysis using autoassociative neural networks (1991)
- Layer Normalization (2016) — link
- Modular Learning in Neural Networks (1987)
- Temporal fusion transformers for interpretable multi-horizon time series forecasting (2021)
- Dimensionality reduction for prediction: Application to Bitcoin and Ethereum (2021) — link
- Deep Switching State Space Model (DS $^ 3$ M) for Nonlinear Time Series Forecasting with Regime Switching (2021) — link
- A regime-switching recurrent neural network model applied to wind time series (2019)
- Markovian RNN: An Adaptive Time Series Prediction Network with HMM-based Switching for Nonstationary Environments (2020)
- Finder: A novel approach of change point detection for multivariate time series (2022) — link
- Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs) (2016)
- The accuracy of asymmetric GARCH model estimation (2019)
- GARCH models: structure, statistical inference and financial applications (2019)
- Sequence to sequence learning with neural networks (2014)
- An inequality and associated maximization technique in statistical estimation for probabilistic functions of Markov processes (1972)
- Long Short-term Memory (1997)
- An Attention Free Transformer (2021)
- Rectified Linear Units Improve Restricted Boltzmann Machines (2010)
- Deep Sparse Rectifier Neural Networks (2010)
- The Vanishing Gradient Problem During Learning Recurrent Neural Nets and Problem Solutions (1998) — link
- Eigenvalues of covariance matrices: Application to neural-network learning (1991) — link
- Integration of paths–A faithful representation of paths by noncommutative formal power series (1958)
- Path signatures for non-intrusive load monitoring (2022)
- Efficient BackProp (2000)
- Enhancing the Locality and Breaking the Memory Bottleneck of Transformer on Time Series Forecasting (2020) — link
- Neural Machine Translation by Jointly Learning to Align and Translate (2016) — link
- On the Properties of Neural Machine Translation: Encoder-Decoder Approaches (2014)
- Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling (2014)
- The perceptron: A probabilistic model for information storage and organization in the brain. (1958)
- A Logical Calculus of Ideas Immanent in Nervous Activity (1943)
- Learning long-term dependencies with gradient descent is difficult (1994) — link
- Markov recurrent neural networks (2018)
- Markovian RNN: An adaptive time series prediction network with HMM-based switching for nonstationary environments (2021)
- Are Transformers Effective for Time Series Forecasting? (2022)
- The Glosten-Jagannathan-Runkle-Generalized Autoregressive Conditional Heteroscedastic Approach to Investigating the Foreign Exchange Forward Premium Volatility (2016)
- Attention Augmented Convolutional Networks (2019) — link
- Generalized autoregressive conditional heteroskedasticity (1986)
- GARCHNET -VALUE-AT-RISK FORECASTING WITH NOVEL APPROACH TO GARCH MODELS BASED ON NEURAL NETWORKS (2021)
- Forecasting the volatility of stock price index: A hybrid model integrating LSTM with multiple GARCH-type models (2018)
- Gold price volatility: A forecasting approach using the Artificial Neural Network–GARCH model (2015)
- Time-dependent series variance learning with recurrent mixture density networks (2013)
- On the Relation between the Expected Value and the Volatility of the Nominal Excess Return on Stocks (1993)
- Kernel Trick for the Cross Section (2019) — link
- Autoencoder asset pricing models (2021)
- Reducing the dimensionality of data with neural networks (2006)
- A Brief Introduction to Path Signatures (2021)
- Neural networks and principal component analysis: Learning from examples without local minima (1989)
- A Comparison of Transformer and LSTM Encoder Decoder Models for ASR (2019) — link
- An Attention Free Long Short-Term Memory for Time Series Forecasting (2022) — link
- Managing Risk in DeFi Portfolios (2022) — link
- Sentiment and bitcoin volatility (2016)
- Realized bitcoin volatility (2017)
- Volatility analysis of bitcoin (2017)
- What can explain the price, volatility and trading volume of Bitcoin? (2019)
- Can volume predict Bitcoin returns and volatility? A quantiles-based approach (2017)
- Clustering algorithms for risk-adjusted portfolio construction (2017)
- Efficient stock portfolio construction by means of clustering (2018)
- Creating diversified portfolios using cluster analysis (2015)
- Topology of correlation-based minimal spanning trees in real and model markets (2003)
- High-frequency cross-correlation in a set of stocks (2001)
- Hierarchical structure in financial markets (1999)
- Deep Fusion of Lead-lag Graphs: Application to Cryptocurrencies (2022) — link
- Some methods for classification and analysis of multivariate observations (1967)
- The Elements of Statistical Learning: Data Mining, Inference, and Prediction (2009)
- Toward maximum diversification (2008)
- Mean-variance analysis in portfolio choice and capital markets (2000)
- Risk reduction by international diversification (1976)
- International portfolio diversification with estimation risk (1985)
- Extracting information from the signature of a financial data stream (2013) — link
- A primer on the signature method in machine learning (2016) — link
- LPSNet: a novel log path signature feature based hand gesture recognition framework (2017)
- Developing the path signature methodology and its application to landmark-based human action recognition (2022)
- Deep signature statistics for likelihood-free time-series models (2021)
- Rough paths, signatures and the modelling of functions on streams (2014) — link
- Differential equations driven by rough signals (1998)
- Uniqueness for the signature of a path of bounded variation and the reduced path group (2010)
- Portfolio Construction Based on Time Series Clustering Method Evidence in the Vietnamese Stock Market (2023)
- A new approach to the economic analysis of nonstationary time series and the business cycle (1989)
- Time series analysis (2020)
- Finite mixture and Markov switching models (2006)
- Multivariate GARCH models: a survey (2006)
- A nonparametric simulated maximum likelihood estimation method (2004)
- Simulated non-parametric estimation of dynamic models (2009)
- Estimation of dynamic latent variable models using simulated non-parametric moments (2012)
- Estimation of dynamic models with nonparametric simulated maximum likelihood (2012)
- Modeling the conditional distribution of interest rates as a regime-switching process (1996)
- Improving GARCH volatility forecasts with regime-switching GARCH (2002)
- A Markov model of unconditional variance in ARCH (1994)
- Detecting bubbles in Bitcoin price dynamics via market exuberance (2021)
- Asymptotic properties of conditional maximum-likelihood estimators (1970)
- Time-varying Hurst exponent for US stock markets (2008)
- Autoregressive conditional heteroskedasticity and changes in regime (1994)
- Theory and inference for a Markov switching GARCH model (2010)
- A new approach to Markov-switching GARCH models (2004)
- Deriving the autocovariances of powers of Markov-switching GARCH models, with applications to statistical inference (2008)
- Maximum likelihood estimation of the Markov-switching GARCH model (2014)
- Markov-switching GARCH models in R: The MSGARCH package (2019)
- On the volatility of cryptocurrencies (2022)
- Forecasting risk with Markov-switching GARCH models: A large-scale performance study (2018)
- Modelling volatility of cryptocurrencies using Markov-Switching GARCH models (2019)
- Is Bitcoin a hedge? How extreme volatility can destroy the hedge property (2022)
- Handbook of matrices (1997)
- Matrix analysis (2012)
- Convergence in distribution of products of random matrices (1984)
- Probability theory: independence, interchangeability, martingales (2012)
- Time Series Forecasting with Transformer Models and Application to Asset Management (2023)
- A comparison of ARIMA and LSTM in forecasting time series (2018)
- Volatility in the cryptocurrency market (2019)
- Unveiling the Influencing Factors of Cryptocurrency Return Volatility (2023)
- Learning from the past, predicting the statistics for the future, learning an evolving system (2013) — link
- The signature of a rough path: uniqueness (2016)
- Cryptocurrency price prediction using tweet volumes and sentiment analysis (2018)
- How Elon Musk’s twitter activity moves cryptocurrency markets (2023)
- Attention-based LSTM for aspect-level sentiment classification (2016)
- At-lstm: An attention-based lstm model for financial time series prediction (2019)
- Mosaic organization of DNA nucleotides (1994)
- Statistical properties of the volatility of price fluctuations (1999)
- Multilayer feedforward networks are universal approximators (1989)
- Neural networks: a comprehensive foundation (1998)
- Approximation by superpositions of a sigmoidal function (1989)
- A survey of transformers (2022)
- Delight: Deep and light-weight transformer (2020) — link
- Exploring the limits of transfer learning with a unified text-to-text transformer (2020)
- The evolved transformer (2019)
- Mask attention networks: Rethinking and strengthen transformer (2021) — link
- Funnel-transformer: Filtering out sequential redundancy for efficient language processing (2020)
- Attentive Language Models Beyond a Fixed-Length Context. Published online June 2, 2019. doi: 10.48550 (1901)
- Megatron-lm: Training multi-billion parameter language models using model parallelism (2019) — link
- Compressive transformers for long-range sequence modelling (2019) — link
- TENER: adapting transformer encoder for named entity recognition (2019) — link
- Tuning multilingual transformers for language-specific named entity recognition (2019)
- Improving multimodal named entity recognition via entity span detection with unified multimodal transformer (2020)
- Learning spatio-temporal transformer for visual tracking (2021)
- Spatio-temporal transformer network for video restoration (2018)
- Generating long sequences with sparse transformers (2019) — link
- Position-aware attention and supervised data improve slot filling (2017)
- Bp-transformer: Modelling long-range context via binary partitioning (2019) — link
- Generating wikipedia by summarizing long sequences (2018) — link
- Linformer: Self-attention with linear complexity (2020) — link
- Poolingformer: Long document modeling with pooling attention (2021)
- Multi-horizon time series forecasting with temporal attention learning (2019)
- Enhancing the locality and breaking the memory bottleneck of transformer on time series forecasting (2019)
- Understanding LSTM–a tutorial into long short-term memory recurrent neural networks (2019) — link
- A Temporal Kolmogorov-Arnold Transformer for Time Series Forecasting (2024) — link
- On the representation of continuous functions of several variables by superpositions of continuous functions of a smaller number of variables (1961)
- KAN: Kolmogorov-Arnold Networks (2024) — link
- Embedding and learning with signatures (2021)
- A signature-based machine learning model for distinguishing bipolar disorder and borderline personality disorder (2018)
- Multiple-output modeling for multi-step-ahead time series forecasting (2010)
- Model specification in multivariate time series (1989)
- Multivariate time series analysis and applications (2018)
- A comparison of direct and iterated multistep AR methods for forecasting macroeconomic time series (2006)
- Time series forecasting using neural networks vs. Box-Jenkins methodology (1991)
- State-space models (1994)
- Dynamic linear models with Markov-switching (1994)
- Vector autoregressive models for multivariate time series (2006)
- Vector autoregressive models (2013)
- Autoregressive convolutional neural networks for asynchronous time series (2018)
- A novel LSTM approach for asynchronous multivariate time series prediction (2019)
- Fast and accurate deep network learning by exponential linear units (elus) (2015) — link
- Language modeling with gated convolutional networks (2017)
- Long short-term memory (1997)
- TKAN: Temporal Kolmogorov-Arnold Networks (2024) — link
- Wav-KAN: Wavelet Kolmogorov-Arnold Networks (2024) — link
- Kolmogorov-arnold networks (kans) for time series analysis (2024) — link
- Temporal pattern attention for multivariate time series forecasting (2019)
- Multivariate temporal convolutional network: A deep neural networks approach for multivariate time series forecasting (2019)
- Informer: Beyond efficient transformer for long sequence time-series forecasting (2021)
- Deep transformer models for time series forecasting: The influenza prevalence case (2020) — link
- Are transformers effective for time series forecasting? (2023)
- Adversarial sparse transformer for time series forecasting (2020)
- A survey on vision transformer (2022)
- Connecting the dots: Multivariate time series forecasting with graph neural networks (2020)
- Multiple convolutional neural networks for multivariate time series prediction (2019)
- Sigmoid-weighted linear units for neural network function approximation in reinforcement learning (2018)
- The vanishing gradient problem during learning recurrent neural nets and problem solutions (1998)
- Backpropagation through time: what it does and how to do it (1990)
- A learning algorithm for continually running fully recurrent neural netwokrs (1989)
- Financial time series forecasting with deep learning: A systematic literature review: 2005–2019 (2020)
- Study of effectiveness of time series modeling (ARIMA) in forecasting stock prices (2014)
- ARMA models and the Box–Jenkins methodology (1997)
- Recurrent neural networks (2001)
- Recurrent neural networks for time series forecasting: Current status and future directions (2021)
- Learning phrase representations using RNN encoder-decoder for statistical machine translation (2014) — link
0 of 220 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.