Paper: arXiv 2306.13661

Authors: Joel Ong, Dorien Herremans

Abstract

A diversified risk-adjusted time-series momentum (TSMOM) portfolio can deliver substantial abnormal returns and offer some degree of tail risk protection during extreme market events. The performance of existing TSMOM strategies, however, relies not only on the quality of the momentum signal but also on the efficacy of the volatility estimator. Yet many of the existing studies have always considered these two factors to be independent. Inspired by recent progress in Multi-Task Learning (MTL), we present a new approach using MTL in a deep neural network architecture that jointly learns portfolio construction and various auxiliary tasks related to volatility, such as forecasting realized volatility as measured by different volatility estimators. Through backtesting from January 2000 to December 2020 on a diversified portfolio of continuous futures contracts, we demonstrate that even after accounting for transaction costs of up to 3 basis points, our approach outperforms existing TSMOM strategies. Moreover, experiments confirm that adding auxiliary tasks indeed boosts the portfolio’s performance. These findings demonstrate that MTL can be a powerful tool in finance.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 8.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced mathematical concepts such as deep neural networks with LSTM modules, multi-task learning frameworks, and volatility scaling equations, indicating high mathematical complexity. It also demonstrates strong empirical rigor by conducting extensive backtests from 2000 to 2020 on futures contracts, accounting for transaction costs, and benchmarking against existing strategies.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Construct robust TSMOM<br>using Deep Multi-Task Learning"] --> B["Methodology<br>Deep Neural Network with<br>Joint Multi-Task Learning"]
  
  B --> C["Input Data<br>Continuous Futures Contracts<br>Jan 2000 - Dec 2020"]
  
  C --> D["Computational Process<br>Simultaneously learn:<br>1. Portfolio Construction<br>2. Auxiliary Volatility Tasks"]
  
  D --> E{"Performance Evaluation<br>Transaction Costs up to 3 bps"}
  
  E -- Comparison --> F["Key Findings/Outcomes"]
  
  F --> G["Superior Performance<br>Outperforms existing TSMOM strategies"]
  F --> H["Robustness<br>Effective even with transaction costs"]
  F --> I["MTL Efficacy<br>Auxiliary tasks boost portfolio performance"]
  F --> J["Risk Protection<br>Offers tail risk protection<br>during extreme events"]

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 multi-task learning time-series momentum model (MTL‑TSM) as a novel deep learning based portfolio optimization method.

Source sentence
We propose a multi-task learning time-series momentum model (MTL‑TSM), which is a novel portfolio optimization method based on deep multi‑task learning.

Assumes 5

  • Transaction costs are set to 3 basis points per trade. execution
    Source sentence
    τ is the transaction cost (set as 3 basis points (bps))
  • All portfolios are volatility‑scaled to a target of 10% annualized volatility. model
    Source sentence
    During our backtest, we applied volatility scaling to all portfolios, bringing the volatility to a target of 10%.
  • The Sharpe ratio is used as the loss function for training the neural network. statistical
    Source sentence
    Since our primary objective when developing the multi‑task model is to construct a portfolio with better risk‑adjusted performance, we employ the Sharpe ratio (Sharpe, 1994), as per Equation (6), as the loss function that is minimized during the training of our neural network model.
  • Features are standardized using a 21‑day sliding‑window z‑score. model
    Source sentence
    Given that these features have varying scales, before using them as inputs to our model, we apply z‑score standardization with a sliding window of length 21 to our features.
Show 1 more
  • The backtest period is from January 2000 to December 2020 on continuous futures contracts. data
    Source sentence
    backtesting from January 2000 to December 2020 on a diversified portfolio of continuous futures contracts

Data

  • Continuous futures contracts portfolio market-prices unknown
    2000-01 to 2020-12 · diversified portfolio of continuous futures contracts
    Source sentence
    backtesting from January 2000 to December 2020 on a diversified portfolio of continuous futures contracts

Recognised sources: Bloomberg Nasdaq Data Link (Quandl)

Costs

Modeled Costs are part of the reported headline results.

Transaction cost set to 3 basis points per trade and incorporated into the return calculation.

Source sentence
τ is the transaction cost (set as 3 basis points (bps))

Finds 4

  • The MTL‑TSMOM strategy outperforms TSMOM and CTA‑MOM by 236 bps and 624 bps respectively over the 20‑year test period. performance
    Source sentence
    In Table 2, the returns of our proposed strategy, which we refer to as MTL‑TSMOM, outperforms both TSMOM and CTA‑MOM by 236 bps and 624 bps, respectively, over the 20 year test period.
  • MTL‑TSMOM achieves an annualized return of 7.90% and Sharpe ratio of 0.81 after transaction costs. performance
    Source sentence
    annualized return of 7.90% for the period of 1st January 2000 to 31st December 2020
  • The strategy exhibits lower long‑term correlation to the US MSCI Total Return Index, making it a strong diversifier. performance
    Source sentence
    our proposed model shows a low correlation to the US MSCI Total Return Index’s returns, thus presenting itself as an extremely attractive diversifier to existing stocks and bonds portfolios.
  • During the Global Financial Crisis and the 2020 Coronavirus Crash, the MTL‑TSMOM outperformed the US MSCI Total Return Index by 57% and 34.4% respectively. performance
    Source sentence
    the US MSCI Total Return Index by 57% and 34.4% in the GFC and Coronavirus Crash of 2020, respectively.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 32 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 3 papers in the archive (2 by independent authors): Explainable Patterns in Cryptocurrency Microstructure (extends it)

Code

https://github.com/joelowj/mtl-tsmom — the authors' own repository

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.