Paper: arXiv 2509.04541
Authors: Kasymkhan Khubiev, Mikhail Semenov, Irina Podlipnova
Abstract
Deep Learning is evolving fast and integrates into various domains. Finance is a challenging field for deep learning, especially in the case of interpretable artificial intelligence (AI). Although classical approaches perform very well with natural language processing, computer vision, and forecasting, they are not perfect for the financial world, in which specialists use different metrics to evaluate model performance. We first introduce financially grounded loss functions derived from key quantitative finance metrics, including the Sharpe ratio, Profit-and-Loss (PnL), and Maximum Draw down. Additionally, we propose turnover regularization, a method that inherently constrains the turnover of generated positions within predefined limits. Our findings demonstrate that the proposed loss functions, in conjunction with turnover regularization, outperform the traditional mean squared error loss for return prediction tasks when evaluated using algorithmic trading metrics. The study shows that financially grounded metrics enhance predictive performance in trading strategies and portfolio optimization.
Complexity vs Empirical Score
- Math Complexity: 6.5/10
- Empirical Rigor: 7.0/10
- Quadrant: Holy Grail — high math complexity, high empirical rigor
Why this score: The paper introduces custom loss functions (e.g., SharpeLoss, MaxDrawDownLoss) with mathematical formulation and experimental validation on real market data, requiring a balance of advanced concepts and backtest implementation.
Research Flowchart
flowchart TD A["Research Goal: Develop interpretable AI for algorithmic trading<br>using financially grounded loss functions"] --> B["Methodology: Propose Financially Grounded Loss Functions<br>Sharpe Ratio, PnL, Max Drawdown"] A --> C["Methodology: Introduce Turnover Regularization<br>Constrain position turnover within limits"] B --> D["Computational Process: Combine loss functions with<br>regularization for training Deep Learning models"] C --> D D --> E["Key Findings: Financially grounded loss functions<br>outperform traditional MSE loss"] D --> F["Key Findings: Enhanced predictive performance<br>in trading strategies & portfolio optimization"]
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 finance‑grounded loss functions (Sharpe‑based, maximum‑drawdown‑based, profit‑and‑loss‑based) and a turnover regularization mechanism to constrain trading activity.
Source sentence
We introduce finance‑grounded loss functions, including Sharpe ratio–based, maximum drawdown–based, and profit-and‑loss–based objectives, that are more suitable for financial time‑series modeling than conventional regression losses.
Assumes 3
- Daily rebalancing with turnover bounded between 30% and 100% per period. execution
Source sentence
The upper bound tb = 1.0 corresponds to a full portfolio turnover per rebalancing period
- Cryptocurrency returns are heavy‑tailed and non‑Gaussian. statistical
Source sentence
The pronounced S‑shaped deviations observed in the QQ plots indicate substantial departures from Gaussianity.
- Historical price data from Binance are accurate and representative of market dynamics. data
Source sentence
We use historical market data obtained from Binance, one of the largest centralized cryptocurrency exchanges (CEXs).
Data
crypto · daily
- Binance cryptocurrency price data market-prices public
Source sentence
Applying these criteria resulted in a universe of N = 61 cryptocurrencies.
Costs
Partial
Turnover regularization is used to limit trading activity, thereby controlling transaction costs, but no explicit cost rates are reported.
Source sentence
To explicitly regulate trading activity, we propose a band turnover regularization that penalizes deviations of the average portfolio turnover from a predefined admissible range.
Finds 2
Sharpe 1.10 · max drawdown 2.2%
- Models trained with finance‑grounded loss functions consistently outperform MSE baselines on algorithmic trading metrics. performance
Source sentence
Empirical results on return prediction tasks demonstrate that models trained with the proposed objectives consistently outperform mean squared error–based baselines when evaluated using algorithmic trading and portfolio-level metrics.
- Turnover regularization improves stability and overall strategy performance compared to classical turnover penalties. performance
Source sentence
explicit turnover regularization not only constrains trading activity within economically meaningful bounds but also enhances stability and overall strategy performance.
Checked by others
No other paper in the archive has been found to support or contradict these findings (compared with 8 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.