Paper: arXiv 2511.08658

Authors: Stanislav Selitskiy

Abstract

We investigate a number of Artificial Neural Network architectures (well-known and more ``exotic’’) in application to the long-term financial time-series forecasts of indexes on different global markets. The particular area of interest of this research is to examine the correlation of these indexes’ behaviour in terms of Machine Learning algorithms cross-training. Would training an algorithm on an index from one global market produce similar or even better accuracy when such a model is applied for predicting another index from a different market? The demonstrated predominately positive answer to this question is another argument in favour of the long-debated Efficient Market Hypothesis of Eugene Fama.

Complexity vs Empirical Score

  • Math Complexity: 2.5/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper applies standard neural network architectures and statistical metrics (MAPE, RMSE) without introducing novel mathematical theory or complex derivations. However, it is heavily data-driven, using a substantial dataset (2005-2022) and rigorous experimental setup including cross-validation and statistical hypothesis testing (Wilcoxon signed-rank), making it backtest-ready and implementation-heavy.

Research Flowchart

  flowchart TD
  A["Research Goal:<br>Does training on one global market index<br>enable accurate prediction of another?"] --> B["Data Inputs:<br>Long-term financial time-series<br>from multiple global market indexes"]
  B --> C["Methodology: Cross-Index Training<br>Train ANN models on Source Market Index<br>(e.g., S&P 500)"]
  C --> D["Computational Process:<br>Transfer Learning / Cross-Market Application<br>Apply trained model to predict<br>Target Market Index (e.g., Nikkei 225)"]
  D --> E["Comparison & Evaluation<br>Measure accuracy vs. models trained<br>specifically on the target index"]
  E --> F{"Key Findings"}
  F --> G["Predominately Positive Results<br>Supports Efficient Market Hypothesis<br>Indices exhibit correlated behavior<br>transferable via Neural Networks"]
  F --> H["Conclusion:<br>Cross-market prediction is viable<br>with ANN architectures"]

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 training ANN models on one market index and cross‑testing them on other global market indexes to assess cross‑training predictive performance.

Source sentence
In our study, ANN models are trained on the index of one single market and then tested on the other three markets, and such experiments are applied to all four markets in a circular fashion.

Assumes 5

  • All potentially available information is immediately incorporated into market prices (Efficient Market Hypothesis). market
    Source sentence
    In various forms of strictness, it postulates that all potentially available information is immediately incorporated into market prices.
  • Cross‑training accuracy can be compared to non‑cross‑training using Wilcoxon signed‑rank tests. statistical
    Source sentence
    Acquired accuracy distributions for cross‑training are subjected to the non‑parametric hypothesis testing Wilcoxon signed rank algorithm to verify that non‑cross‑training and cross‑training accuracy distribution either have no shift or “right” (greater) shift for non‑cross‑training distributions
  • Mean Absolute Percentage Error (MAPE) and Root Mean Square Error (RMSE) are appropriate accuracy metrics for the forecasts. model
    Source sentence
    As an accuracy metrics, we use the Mean Absolute Percentage Error (MAPE)
  • No averaging or use of data from multiple markets for training is performed. data
    Source sentence
    No averaging or other use of data from multiple markets for training is done because the study aims to experimentally confirm the alleged by EMH synchronicity of the markets
Show 1 more
  • The four selected indexes (NASDAQ, Dow Jones, NIKKEI, DAX) are sufficient to limit combinatorial complexity while representing global markets. other
    Source sentence
    The reason for the particular indexes selection is an attempt to limit the combinatorial complexity of the experiments (only four indexe…

Data

  • NASDAQ index · tradingeconomics.com market-prices public
    beginning of 2005 to end of January 2022
    Source sentence
    The data for NASDAQ, Dow Jones, NIKKEI, and DAX indexes for the beginning of 2005 year up to the end of January 2022 were used in computational experiments. The data were downloaded from https://tradingeconomics.com.
  • Dow Jones index · tradingeconomics.com market-prices public
    beginning of 2005 to end of January 2022
    Source sentence
    The data for NASDAQ, Dow Jones, NIKKEI, and DAX indexes for the beginning of 2005 year up to the end of January 2022 were used in computational experiments. The data were downloaded from https://tradingeconomics.com.
  • NIKKEI index · tradingeconomics.com market-prices public
    beginning of 2005 to end of January 2022
    Source sentence
    The data for NASDAQ, Dow Jones, NIKKEI, and DAX indexes for the beginning of 2005 year up to the end of January 2022 were used in computational experiments. The data were downloaded from https://tradingeconomics.com.
  • DAX index · tradingeconomics.com market-prices public
    beginning of 2005 to end of January 2022
    Source sentence
    The data for NASDAQ, Dow Jones, NIKKEI, and DAX indexes for the beginning of 2005 year up to the end of January 2022 were used in computational experiments. The data were downloaded from https://tradingeconomics.com.

Costs

Not applicable The paper reports no trading or execution result.

Finds 4

  • Cross‑trained models achieve consistently better accuracy for the NIKKEI index across all ANN models. predictive-accuracy
    Source sentence
    cross‑trained accuracy being consistently better across all ANN models for NIKKEI index
  • Cross‑trained models perform consistently worse for the Dow Jones index across all ANN models. predictive-accuracy
    Source sentence
    consistently worse across all ANN models for Dow Jones
  • When a shift bound of 0.02 is applied, all models and stocks show at least no worse accuracy for cross‑trained models compared to non‑cross‑trained ones. statistical
    Source sentence
    If the shift is bounded by 0.02 (which is less than the standard deviation of even the best and narrowest accuracy distributions), then all models and stocks demonstrate at least no worse accuracy of the cross‑trained models compared to the ablation non‑cross‑trained ones.
  • The results support a weak form of the Efficient Market Hypothesis, indicating common patterns across global markets. theoretical
    Source sentence
    the study supports a weak form of EMH, indirectly supporting the idea of existing of the patterns common for global markets

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 64 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

https://github.com/Selitskiy/LTTS — 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.

Cited References (41)

  1. The Elements of Statistical Learning (2001)
  2. On the representation of continuous functions of several variables by superpositions of continuous functions of a smaller number of variables (1961)
  3. Representation properties of networks: Kolmogorov’s theorem is irrelevant (1989)
  4. Kolmogorov’s Theorem Is Relevant (1991)
  5. Approximation theory of the MLP model in neural networks (1999)
  6. Polynomial theory of complex systems (1971)
  7. The GMDH algorithm of Ivakhnenko (1981)
  8. Initial conditions and boundary conditions (1994)
and 33 more
  1. Searching for Activation Functions (2017) — link
  2. Artificial Neural Networks as a Tool for Recognition of Movements by Electroencephalograms. (2018)
  3. Universal Approximation Using Radial-Basis-Function Networks (1991)
  4. New RBF neural network classifier with optimized hidden neurons number (2004)
  5. Radial basis functions, multi-variable functional interpolation and adaptive networks (1988)
  6. Going deeper with convolutions (2015)
  7. Deep residual learning for image recognition (2016)
  8. Kolmogorov’s Gate Non-linearity as a Step toward Much Smaller Artificial Neural Networks (2022)
  9. Efficient capital markets: A review of theory and empirical work (1970)
  10. Efficient Market Hypothesis (EMH): Past, Present and Future (2008)
  11. History of the efficient market hypothesis (2011)
  12. Inefficient markets: An introduction to behavioural finance (2000)
  13. Orange juice and weather (1984)
  14. The performance of mutual funds in the period 1945-1964 (1968)
  15. International transmission of stock market movements (1989)
  16. Mean reversion in stock prices: Evidence and implications (1988)
  17. What every CFO should know about scientific progress in financial economics: What is known and what remains to be resolved (1994)
  18. Neural networks and statistical techniques in marketing research: A conceptual comparison (1994)
  19. Deep learning for stock market prediction (2020)
  20. Deep architectures for long-term stock price prediction with a heuristic-based strategy for trading simulations (2019)
  21. A neural network ensemble method with jittered training data for time series forecasting (2007)
  22. Self-supervised contrastive pre-training for time series via time-frequency consistency (2022)
  23. Mixing up contrastive learning: Self-supervised representation learning for time series (2022)
  24. Recurrent neural networks for financial time-series modelling (2018)
  25. Multi-source transfer learning with ensemble for financial time series forecasting (2020)
  26. Deep learning for biometric face recognition: experimental study on benchmark data sets (2020)
  27. Evolving polynomial neural networks for detecting abnormal patterns (2016)
  28. Extraction of Texture Features from X-Ray Images: Case of Osteoarthritis Detection (2018)
  29. Bayesian assessment of newborn brain maturity from two-channel sleep electroencephalograms (2012)
  30. Informativeness of sleep cycle features in Bayesian assessment of newborn electroencephalographic maturation (2011)
  31. Elements of Active Continuous Learning and Uncertainty Self-awareness: A Narrow Implementation for Face and Facial Expression Recognition (2023)
  32. HYBRID CONVOLUTIONAL-MULTILAYER PERCEPTRON ARTIFICIAL NEURAL NETWORK FOR PERSON RECOGNITION BY HIGH GAMMA EEG FEATURES (2022)
  33. Using Statistical and Artificial Neural Networks Meta-learning Approaches for Uncertainty Isolation in Face Recognition by the Established Convolutional Models (2022)

0 of 41 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.