Paper: arXiv 2309.10546

Authors: Jakub Michańków, Paweł Sakowski, Robert Ślepaczuk

Abstract

This paper investigates the issue of an adequate loss function in the optimization of machine learning models used in the forecasting of financial time series for the purpose of algorithmic investment strategies (AIS) construction. We propose the Mean Absolute Directional Loss (MADL) function, solving important problems of classical forecast error functions in extracting information from forecasts to create efficient buy/sell signals in algorithmic investment strategies. Finally, based on the data from two different asset classes (cryptocurrencies: Bitcoin and commodities: Crude Oil), we show that the new loss function enables us to select better hyperparameters for the LSTM model and obtain more efficient investment strategies, with regard to risk-adjusted return metrics on the out-of-sample data.

Complexity vs Empirical Score

  • Math Complexity: 4.0/10
  • Empirical Rigor: 6.0/10 (capped from 7.0: transaction costs not modelled; see penalty flags)
  • Quadrant: Street Traders — empirical and applied, lower math complexity (effective empirical rigor capped from 7.0)

Why this score: The paper presents a novel loss function (MADL) with a defined mathematical formula but lacks deep theoretical derivations or proofs, keeping math complexity moderate. It demonstrates empirical rigor by applying the method to real financial data (Bitcoin and Crude Oil) using walk-forward optimization and evaluating out-of-sample risk-adjusted returns, making it backtest-ready.

Research Flowchart

  flowchart TD
  Start(["Research Goal<br>Develop an effective loss function<br>for algorithmic investment strategies"]) --> Data["Input Data<br>BTC & Crude Oil Historical Data"]
  Data --> Model["LSTM Model Architecture"]
  Model --> Comp["Computational Process<br>Train with:<br>1. MAE Loss<br>2. Proposed MADL Loss"]
  Comp --> Select["Hyperparameter Selection<br>Optimize based on loss function"]
  Select --> Strat["Generate Algorithmic Investment Strategy<br>Trading Signals (Buy/Sell)"]
  Strat --> Eval["Out-of-Sample Evaluation<br>Compare Risk-Adjusted Returns"]
  Eval --> Out["Key Findings<br>MADL enables better hyperparameter selection<br>and more efficient investment strategies"]

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 new loss function called Mean Absolute Directional Loss (MADL) for optimizing machine learning models in algorithmic investment strategies.

Source sentence
We propose the Mean Absolute Directional Loss (MADL) function

Assumes 2

  • The study uses simple daily returns of the two assets as the basis for modeling. data
    Source sentence
    We decided to use simple returns based on daily data of two assets from various asset classes: Bitcoin (BTC) and Crude Oil (represented by ETF named USO).
  • Minimizing the MADL loss will produce a profitable strategy when the loss value is negative and a loss when it is positive. model
    Source sentence
    In our model, this function is minimized, ensuring that the strategy will produce a profit if it returns negative values and loss if it returns positive values.

Data

  • Bitcoin (BTC) price data · Kraken, Bitfinex, BTC-e, CEX, Coinbase exchanges market-prices public
    April 1, 2023 to December 31, 2021 · Bitcoin
    Source sentence
    Source of the BTC data: Kraken, Bitfinex, BTC-e, CEX, and Coinbase exchanges
  • Crude Oil ETF (USO) price data · stooq.com market-prices public
    April 1, 2023 to December 31, 2021 · USO
    Source sentence
    USO data was obtained from stooq.com.

Recognised sources: Bitfinex Coinbase Kraken

Costs

Not modeled A trading or execution result is reported without costs.

Finds 3

  • Using the MADL loss function leads to better hyperparameter selection for the LSTM model and more efficient investment strategies. performance
    Source sentence
    the new loss function enables us to select better hyperparameters for the LSTM model and obtain more efficient investment strategies
  • The MADL/MADL approach consistently yields the highest risk‑adjusted return metrics among the tested loss‑function combinations. performance
    Source sentence
    the result is always the best in s of MADL/MADL approach.
  • Regression results show that the MADL/MADL strategy has a significantly higher alpha (0.0029) and t‑stat (4.027) compared to other configurations. statistical
    Source sentence
    MADL/MADL 0.0029 0.001 4.027 0.000

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 21 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 6 papers in the archive

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.