Paper: arXiv 2307.13546
Authors: Haoyang Cao, Haotian Gu, Xin Guo, Mathieu Rosenbaum
Abstract
In this work, we explore the possibility of utilizing transfer learning techniques to address the financial portfolio optimization problem. We introduce a novel concept called “transfer risk”, within the optimization framework of transfer learning. A series of numerical experiments are conducted from three categories: cross-continent transfer, cross-sector transfer, and cross-frequency transfer. In particular, 1. a strong correlation between the transfer risk and the overall performance of transfer learning methods is established, underscoring the significance of transfer risk as a viable indicator of “transferability”; 2. transfer risk is shown to provide a computationally efficient way to identify appropriate source tasks in transfer learning, enhancing the efficiency and effectiveness of the transfer learning approach; 3. additionally, the numerical experiments offer valuable new insights for portfolio management across these different settings.
Complexity vs Empirical Score
- Math Complexity: 8.0/10
- Empirical Rigor: 6.0/10 (capped from 6.5: transaction costs not modelled; see penalty flags)
- Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 6.5)
Why this score: The paper introduces a novel mathematical concept (transfer risk) and references advanced theoretical frameworks from transfer learning, including various divergences and optimal transport, indicating high mathematical complexity. While it lacks code snippets or direct backtest metrics, the detailed numerical experiments across three distinct transfer categories (cross-continent, sector, and frequency) with specific performance insights and correlation analysis demonstrate significant data-driven implementation and empirical validation.
Research Flowchart
flowchart TD A["Research Goal: Utilize Transfer Learning for Portfolio Optimization"] --> B["Key Concept: Define 'Transfer Risk' Metric"] B --> C["Data & Methodology: Cross-Continent, Sector & Frequency Transfers"] C --> D["Computational Process: Evaluate Transfer Risk vs. Portfolio Performance"] D --> E["Key Finding 1: Strong Correlation between Transfer Risk & Performance"] D --> F["Key Finding 2: Transfer Risk Efficiently Identifies Best Source Tasks"] D --> G["Key Finding 3: New Insights for Portfolio Management"]
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 a novel concept called transfer risk for portfolio optimization via transfer learning.
Source sentence
concept called “transfer risk”, within the optimization framework for transfer learning
Assumes 5
- Source and target tasks share the same input and output spaces (Rd and R). model
Source sentence
assume that the source and target tasks S and T share the same input and output spaces denoted by XS = XT = Rd and YS = YT = R
- Both source and target models are linear mappings from returns to portfolio return. model
Source sentence
the admissible sets of source and target models are restricted to: AS = AT = {f : Rd →R|f(r) = ϕ⊤r for some ϕ ∈X}
- The regularization hyper‑parameter λ is positive and controls the strength of the L2 penalty. model
Source sentence
λ > 0 is a hyper-parameter controls the power of the regularization
- Over‑night holding is not allowed for trading frequencies ranging from 1‑minute to 130‑minute. execution
Source sentence
we assume that over-night holding is not allowed for trading frequencies ranging from 1-minute to 130-minute
Show 1 more
- Historical daily returns are used to estimate mean and covariance matrices for both source and target tasks. data
Source sentence
Empirically, µP and ΣP are estimated from the historical return
Data
- Source training data (cross‑continent transfer) market-prices unknown
Source sentence
Source training data: it consists of the daily returns of ten source assets, from February 2000 to February 2020.
- Target training data (cross‑continent transfer) market-prices unknown
Source sentence
Target training data: it consists of the daily returns of ten target assets, from February 2015 to February 2020.
- Target testing data (cross‑continent transfer) market-prices unknown
Source sentence
Target testing data: it consists of the daily returns of ten target assets, from February 2020 to September 2021.
- Sector transfer data market-prices unknown
Source sentence
Target training data: it consists of the daily returns of ten stocks from a given target sector, from February 2015 to February 2020.
- Frequency transfer data market-prices unknown
Source sentence
it consists of the returns of ten stocks, sampled under a given target frequency, from September 2019 to February 2020.
Costs
Not modeled
Finds 5
- A strong correlation is observed between transfer risk and transfer learning performance across markets. statistical
Source sentence
we observe a strong correlation between transfer risk and the transfer learning performance across all the different markets.
- Negative correlation between transfer risk scores and Sharpe ratios; lower transfer risk leads to higher Sharpe. performance
Source sentence
In general, the negative correlation between Sharpe ratios and scores is observed
- For the Health Care target sector, source sectors Information Technology, Health Care, and Consumer Discretionary yield higher Sharpe ratios, while Energy is unsuitable. performance
Source sentence
Information Technology, Health Care, and Consumer Discretionary are desirable source sectors when the target sector is Health Care, while Energy is not a suitable choice.
- Source frequencies of 65‑minute and 130‑minute provide better transfer learning performance than higher frequencies. performance
Source sentence
Figure 7 reveal that 65-minute and 130-minute are more appropriate candidates for the source frequency, since they lead to much better transfer learning performance
- Transfer risk can be computed efficiently and used as a prior estimate of final learning outcome before full transfer learning. theoretical
Source sentence
transfer risk is an easy-to-compute and viable quantity as a prior estimate of the final learning outcome
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 17 related papers). Coverage is partial, so this is neither replication nor refutation.
2 papers in the archive : Portfolio Optimization via Transfer Learning
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.