Paper: arXiv 2312.16448

Authors: Erdinc Akyildirim, Matteo Gambara, Josef Teichmann, Syang Zhou

Abstract

We present convincing empirical results on the application of Randomized Signature Methods for non-linear, non-parametric drift estimation for a multi-variate financial market. Even though drift estimation is notoriously ill defined due to small signal to noise ratio, one can still try to learn optimal non-linear maps from data to future returns for the purposes of portfolio optimization. Randomized Signatures, in contrast to classical signatures, allow for high dimensional market dimension and provide features on the same scale. We do not contribute to the theory of Randomized Signatures here, but rather present our empirical findings on portfolio selection in real world settings including real market data and transaction costs.

Complexity vs Empirical Score

  • Math Complexity: 7.5/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced mathematical concepts from rough path theory and randomized signatures, involving non-linear drift estimation and high-dimensional feature mapping. It is evaluated on real market data with transaction costs and compares against standard benchmarks, indicating strong empirical validation and implementation focus.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Apply Randomized Signatures<br>to Optimal Portfolio Selection"] --> B["Data Inputs<br>Real Market Data: Equities<br>Including Transaction Costs"]
  B --> C["Methodology:<br>Randomized Signature Methods"]
  C --> D["Computational Process:<br>Non-linear, Non-parametric<br>Drift Estimation"]
  D --> E["Outcome:<br>Learn Optimal Maps<br>for Future Returns"]
  E --> F["Key Finding:<br>Effective Portfolio Selection<br>in Real-World Settings"]

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 using Randomized Signature Methods as a non‑linear, non‑parametric drift estimator for portfolio selection.

Source sentence
We present convincing empirical results on the application of Randomized Signature Methods for non-linear, non-parametric drift estimation for a multi-variate financial market.

Assumes 6

  • Preferences of the investor are known. model
    Source sentence
    In our context, we assume that the preferences are known and the stochastic process S is determined by an Itˆo semi-martingale with continuous trajectories
  • Asset price dynamics follow an Itô semi‑martingale with continuous trajectories. market
    Source sentence
    we let [−Tobs, 0] to be the observation period and (s 7−→Xs)s∈[−Tobs,0] to be the observed data trajectory.
  • Drift and volatility functions are unknown and path‑dependent. model
    Source sentence
    once all factor processes are considered, the quadratic covariation is calculated as follows...
  • A mild martingale assumption on the stochastic integral for drift estimation. statistical
    Source sentence
    Next we consider the task of drift estimation (under a mild martingale assumption on the stochastic integral):
Show 2 more
  • No short‑selling and maximum weight per asset limited to 20%. execution
    Source sentence
    we impose holding constraints on the weights by not allowing short selling and additionally limiting the maximum allowed weight per single asset at 20% of the total assets.
  • Transaction costs are ignored for simulated‑data experiments. execution
    Source sentence
    For the sake of simplicity, we do not consider transaction costs for our experiments using simulated data

Data

  • S&P 500 data market-prices unknown
    S&P 500 constituents
    Source sentence
    In our experiments, we choose rm ∈{0, 0.05, 0.1}, rv ∈{0.01, 0.03, 0.05, 0.3, 1.0} and rd ∈{50, 60, 70, 100}, respectively.
  • Synthetic simulations synthetic unknown
    Source sentence
    In this section we present the empirical results from our algorithm with two different portfolio construction methodologies using both real and simulated prices

Costs

Partial Costs appear only as a robustness check, a flat fee without spread or impact, or outside the headline numbers.

Four cost levels are examined (0%, 0.3%, 0.6%, 1%) with hyper‑parameter tuning; costs are applied only in real‑world data experiments.

Source sentence
Figure 9: Performance under four different transaction costs, namely {0%, 0.3%, 0.6%, 1%}.

Finds 5

  • The Randomized Signature strategy out‑performed the 1/n benchmark on monthly returns in 67.28% of the months in simulated data. performance
    Source sentence
    if we consider the 1/n strategy as a possible competitor, from Figure 4 we observe that our methodology over-performed the former 67.28% of the times on monthly returns.
  • Across the full real‑world period, the proposed portfolio out‑performed all benchmark portfolios for all transaction‑cost levels. performance
    Source sentence
    Overall, one can see that our strategy outperforms the benchmark portfolio on most of the time segments and trading cost levels.
  • Even with the worst hyper‑parameter configuration, the method delivered higher annualised return and Sharpe ratio than benchmarks. performance
    Source sentence
    Even though the results vary significantly over the hyper-parameter space used for our experiments, we show that our portfolio over-performs for both annualized return and annualized Sharpe ratio even in the worst hyper-parameter configuration.
  • When signal‑to‑noise ratio falls below a lower bound, all predictions break down and the 1/n portfolio becomes best. negative
    Source sentence
    It also suggests that there is a lower bound for the signal-to-noise ratio under which all predictions break down and the one-over-n portfolio becomes the best-performing portfolio.
  • The method is robust to transaction costs; performance degrades gracefully as costs increase. performance
    Source sentence
    seen, for the environment without transaction costs, the best hyper-parameter configurations are in the upper left corner, whereas with increasing transaction costs, the best-performing configurations move more and more to the lower right corner.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 26 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 4 papers in the archive (4 by independent authors): Universal randomised signatures for generative time series … (uses its method or data); Kernel Learning for Mean-Variance Trading Strategies (uses its method or data)

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 (89)

  1. Duality theory for portfolio optimisation under transaction costs (2016) — link
  2. Encoding of high-frequency order information and prediction of short-term stock price by deep learning (2019) — link
  3. Internet Appendix for Deep Learning in Asset Pricing (2020) — link
  4. Forecasting trends with asset prices (2017) — link
  5. Deep Learning for Portfolio Optimization — link
  6. Mathematical Finance: A Very Short Introduction (2019) — link
  7. Far from the madding crowd: collective wisdom in prediction markets (2019) — link
  8. In Defense of Portfolio Optimization: What If We Can Forecast? (2019) — link
and 81 more
  1. Mean-variance portfolio optimization when means and covariances are unknown (2011) — link
  2. Machine Learning for Asset Managers (2020) — link
  3. A neural network enhanced volatility component model (2020) — link
  4. Constrained portfolio optimisation: The state-of-the-art Markowitz models (2016) — link
  5. DEEP NO-ARBITRAGE ASSET PRICING (2020)
  6. Deep neural network framework based on backward stochastic differential equations for pricing and hedging American options in high dimensions (2020) — link
  7. XVA analysis from the balance sheet (2020) — link
  8. Neural network-based automatic factor construction (2020) — link
  9. Practical Bayesian support vector regression for financial time series prediction and market condition change detection (2017) — link
  10. Stock market trend prediction using a functional time series approach (2020) — link
  11. Deep hedging (2019) — link
  12. Randomized Signature Layers for Signal Extraction in Time Series Data (2022) — link
  13. Optimizing a portfolio of mean-reverting assets with transaction costs via a feedforward neural network (2020) — link
  14. In defense of optimization: The fallacy of 1/N (2010) — link
  15. Computing efficient frontiers using estimated parameters (1993) — link
  16. Price signatures (2019) — link
  17. Algorithmic trading in a microstructural limit order book model (2020) — link
  18. Deep learning for limit order books (2019) — link
  19. Does Academic Research Destroy Stock Return Predictability? (2016) — link
  20. Empirical Asset Pricing via Machine Learning (2020) — link
  21. Advances in Financial Machine Learning (2020) — link
  22. Performance Analysis of Portfolio Optimisation Strategies: Evidence from the Exchange Market (2017) — link
  23. Structural asset pricing theory with wavelets (2019) — link
  24. Risk discriminating portfolio optimization (2018) — link
  25. Forecasting jump arrivals in stock prices: new attention-based network architecture using limit order book data (2019) — link
  26. On portfolio optimization: Imposing the right constraints (2013) — link
  27. Enhancing the momentum strategy through deep regression (2019) — link
  28. Deep learning volatility: a deep neural network perspective on pricing and calibration in (rough) volatility models (2020) — link
  29. Optimal versus naive diversification: How inefficient is the 1/N portfolio strategy? (2009) — link
  30. Stochastic automatic differentiation: automatic differentiation for Monte-Carlo simulations (2019) — link
  31. Nonlinear shrinkage of the covariance matrix for portfolio selection: Markowitz meets goldilocks (2017) — link
  32. Deep learning for ranking response surfaces with applications to optimal stopping problems (2020) — link
  33. A Random Walk Down Wall Street: The Time-Tested Strategy For Successful Investing (2017) — link
  34. The fundamental law of active management: Redux (2017) — link
  35. Deep learning in asset pricing (2019) — link
  36. Learning from the past, predicting the statistics for the future, learning an evolving system (2016) — link
  37. nlshrink: Non-Linear Shrinkage Estimation of Population Eigenvalues and Covariance Matrices (2016) — link
  38. Mutual Fund Performance (1966)
  39. Extracting information from the signature of a financial data stream (2013) — link
  40. Simple heresy: Rules of thumb challenge complex financial analyses (2011) — link
  41. The Impact of Proportional Transaction Costs on Systematically Generated Portfolios (2020) — link
  42. Lectures on the theory of estimation of many parameters (1986)
  43. A Century of Evidence on Trend-Following Investing (2017)
  44. Profitability of Momentum Strategies: An Evaluation of Alternative Explanations (2001)
  45. Derivatives pricing using signature payoffs (2018) — link
  46. Numerical Method for Model-free Pricing of Exotic Derivatives in Discrete Time Using Rough Path Signatures (2019) — link
  47. Non-parametric pricing and hedging of exotic derivatives (2020)
  48. Signature Trading: A Path-Dependent Extension of the Mean-Variance Framework with Exogenous Signals — → 2308.15135
  49. A Monte Carlo method for optimal portfolios (2003)
  50. Benchmarking, portfolio insurance and technical analysis: a Monte Carlo comparison of dynamic strategies of asset allocation (2003)
  51. Multi-period mean–variance portfolio optimization based on Monte-Carlo simulation (2016)
  52. Combining equilibrium, resampling, and analyst’s views in portfolio optimization (2012)
  53. Stochastic portfolio theory (2002)
  54. Derivatives pricing using signature payoffs (2018) — link
  55. Numerical method for model-free pricing of exotic derivatives in discrete time using rough path signatures (2019)
  56. Discrete-time signatures and randomness in reservoir computing (2021)
  57. Expressive power of randomized signature (2021)
  58. A Primer on the Signature Method in Machine Learning (2016) — link
  59. Portfolio selection: Efficient diversification of investments (1959)
  60. Market value, time, and risk (1961)
  61. Toward a theory of market value of risky assets (1961)
  62. Capital asset prices: A theory of market equilibrium under conditions of risk (1964)
  63. The valuation of risk assets and the selection of risky investments in stock portfolios and capital budgets (1975)
  64. Equilibrium in a capital asset market (1966)
  65. Global portfolio optimization (1992)
  66. Factor-based portfolio optimization (2023)
  67. Risk parity portfolios with risk factors (2016)
  68. Risk parity portfolio optimization under a Markov regime-switching framework (2019)
  69. Least-squares approach to risk parity in portfolio selection (2016)
  70. Signature Methods in Stochastic Portfolio Theory — → 2310.02322
  71. Machine learning and portfolio optimization (2018)
  72. Deep learning for finance: deep portfolios (2017)
  73. Robustness in Portfolio Optimization (2023)
  74. A generalised signature method for multivariate time series feature extraction (2020) — link
  75. Optimal stopping with signatures (2023)
  76. Applications of Signature Methods to Market Anomaly Detection (2022) — link
  77. The Markowitz Optimization Enigma: Is ‘Optimized’ Optimal? (1989)
  78. Volatility is (mostly) path-dependent (2023) — link
  79. Extensions of Lipschitz maps into a Hilbert space (1984) — link
  80. Building portfolios based on machine learning predictions (2022) — link
  81. Algorithm 1004: The iisignature Library: Efficient Calculation of Iterated-Integral Signatures and Log Signatures (2020)

2 of 89 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.