The core insight behind FTSformer is deceptively simple but technically demanding. Most existing forecasting architectures, including sophisticated transformer models that have dominated long-horizon time series benchmarks, concentrate almost exclusively on the historical behavior of the target variable. They learn the internal dynamics of a single asset and extrapolate from them. Yet financial time series are shaped by an intricate web of external drivers: macroeconomic indicators, sector indices, and cross-asset signals all exert latent coupling effects on the target. When these exogenous variables are ignored or treated as generic side inputs, models lose critical information about why a market moves, not merely how it moved. The researchers set out to build an architecture that simultaneously captures cross-asset interactions, multi-scale temporal dynamics, and the notoriously complex loss landscape that deep forecasting models must navigate.

At the heart of the framework lies the Dual-Domain Fusion Layer, or DDF, a module designed to extract and fuse heterogeneous exogenous information in both the time domain and the frequency domain. Time-domain features capture raw temporal patterns such as trends and sudden shocks, while frequency-domain representations expose periodicities and oscillatory components that are invisible in raw waveforms. By jointly processing both domains and explicitly emphasizing their intrinsic heterogeneity, the DDF layer allows the model to distinguish, for example, between a slow-moving macroeconomic drift and a fast, cyclical sector rotation that may both influence the same stock. This dual-domain strategy reflects a broader trend in modern time series research, where frequency-enhanced architectures such as FEDformer have shown that spectral information can substantially improve long-term forecasting, but extends it by systematically integrating external signals rather than only the target series.

Complementing the fusion layer, the researchers introduce a Multi-Scale Temporal Partition mechanism, abbreviated MSTP, together with a Channel-Modulation Preprocessor known as CMP. Financial data unfold at multiple characteristic timescales simultaneously: intraday noise, weekly rhythms, monthly cycles, and secular trends all coexist in a single price series. The MSTP mechanism partitions the input sequence across these scales so that the model can attend to short-lived fluctuations and long-range dependencies without forcing one convolutional or attention window to serve all purposes. The CMP, in turn, addresses a practical problem that plagues multivariate forecasting: different input channels arrive with vastly different signal strengths. A dominant channel can overwhelm weaker but informative ones during training, destabilizing gradients and slowing convergence. The CMP dynamically adapts the weights assigned to each input channel, balancing signal strengths before the core temporal modeling module ever sees the data. According to the authors, this preprocessing step measurably enhances both training stability and convergence speed, an advantage that becomes evident in the ablation experiments reported in the paper.

The temporal backbone of FTSformer is a component the authors call BA-LSTM, a bidirectional long short-term memory architecture tasked with capturing both short-term market fluctuations and long-term dependencies. LSTMs, first popularized in the deep learning era for sequence modeling, remain a powerful tool for financial prediction, and hybrid designs combining recurrent networks with transformers have proliferated in recent literature. In FTSformer, the BA-LSTM operates in concert with the transformer’s attention machinery, providing a recurrent pathway that preserves order-sensitive information while the attention layers model global interactions across the multi-scale, exogenously enriched representations. This division of labor echoes a recurring theme in the forecasting literature, where purely attention-based models sometimes struggle with the strong nonlinearity and regime changes characteristic of financial data.

The empirical evaluation is unusually broad for a methods paper in this domain. The team tested FTSformer on multiple public benchmark datasets as well as a real-world Canadian traffic dataset, deliberately stepping outside finance to probe cross-domain transferability. The headline result is a reduction in forecasting errors of up to 13.4 percent on the Exchange dataset, a benchmark notorious for its nonstationary exchange-rate dynamics that punish rigid models. Across the full suite of public benchmarks, the model achieved average reductions of 1.54 percent in root mean square error and 2.22 percent in mean absolute error over the strongest baseline. While single-digit average gains may appear modest, in competitive forecasting benchmarks where state-of-the-art models are separated by fractions of a percent, such improvements are meaningful, and the consistency of gains across heterogeneous datasets suggests the architecture captures something genuinely general rather than dataset-specific.

Equally notable are the results on training stability and robustness. The authors report that FTSformer substantially outperforms existing baselines not only in accuracy but also in the stability of its training process, an attribute they attribute to the Channel-Modulation Preprocessor’s balancing of channel signals before temporal modeling. They further evaluated robustness to exogenous perturbations, deliberately injecting noise or distortions into the external variables to simulate real-world conditions in which exogenous data sources may be incomplete, delayed, or corrupted. The model’s performance degraded more gracefully than that of competing methods, reinforcing the value of explicitly modeling the heterogeneity of external information rather than naively concatenating it into the input tensor. This robustness finding carries practical weight for financial institutions, where data pipelines are rarely pristine and a model that collapses when an indicator feed fails is of limited operational use.

The study situates itself within a rapidly evolving landscape of time series transformers. The original transformer architecture, introduced in 2017 for machine translation, was adapted to forecasting by models such as Informer and Autoformer, which introduced sparse attention and decomposition strategies for long sequences. More recently, iTransformer reoriented the attention mechanism across variables rather than time steps, while TimeXer explicitly empowered transformers with exogenous variables, a direct conceptual ancestor of FTSformer’s exogenous fusion philosophy. Meanwhile, finance-specific transformers such as Master have explored market-guided attention, and questions about whether transformers are even the right inductive bias for time series, posed pointedly by Zeng and colleagues in their DLinear work, continue to animate the field. FTSformer’s contribution is to combine the multi-scale decomposition trend with a principled, dual-domain treatment of exogenous heterogeneity, targeting the specific pathologies of financial data: high volatility, strong nonlinearity, and pronounced dependence on external drivers.

The authors are careful to frame the work as a flexible, exogenous-aware forecasting framework rather than a finance-only tool. The inclusion of the Canadian traffic dataset is a deliberate demonstration that the architecture’s benefits extend beyond markets, and the paper reports promising cross-domain transferability, suggesting that the dual-domain fusion and channel modulation principles could generalize to energy demand, weather, epidemiology, or any multivariate setting where external covariates drive the target. For practitioners in quantitative finance, the framework arrives amid growing interest in machine learning methods for asset pricing and risk management, following influential work such as Gu, Kelly, and Xiu’s empirical asset pricing study and the continuing proliferation of deep learning surveys in the field. The research was supported by the National Natural Science Foundation of China under Grant 62103177 and by the Key Technology Research and Development Program of Shandong Province under Grant 2025TSGCCZZB0893.

What makes the paper resonant beyond its benchmark numbers is its conceptual reframing of the forecasting problem. Rather than asking how best to extrapolate a single series, FTSformer treats prediction as an exercise in multi-source reasoning: the target’s own multi-scale history, the time-domain and frequency-domain signatures of external drivers, and the relative reliability of every input channel must all be weighed together, dynamically, at every forecasting step. As markets grow more interconnected and data feeds multiply, that framing is likely to influence how the next generation of forecasting systems is built. The study, received on 2 February 2026, accepted on 19 August 2026, and published on 30 August 2026, adds a technically detailed and empirically validated entry to that conversation, one that argues the future of accurate forecasting lies not in ever-larger single-series models but in architectures that know how to listen to the world around the signal.

Radfar’s research first addresses the widespread use of Long Short-Term Memory (LSTM) networks in financial time series prediction—a method extensively employed due to its reputed ability to grasp temporal dependencies. The paper rigorously critiques prior works that built on LSTM’s apparent successes, revealing that many claims overstate the model’s real-world effectiveness. Specifically, the study demonstrates how LSTM models, often trained on limited datasets, fail to translate their apparent predictive power when applied to realistic trading environments, leading to misguided expectations among both practitioners and academic circles.

Moving beyond the LSTM paradigm, the study explores two alternative deep learning architectures: transformers and convolutional neural networks (CNNs). These models were chosen for their architectural differences and strengths—the transformer’s capacity for capturing long-range dependencies through attention mechanisms, and CNN’s prowess in identifying local features via convolutional filters. Experimental results reveal that these architectures indeed outperform day-to-day LSTM models in standard forecast accuracy benchmarks. However, an intriguing and somewhat disquieting observation emerged; these refined networks generated forecasts that were largely agnostic to specific historical price movements over the preceding 100 days.

Instead of leveraging nuanced past price changes for predictions, the models gravitated toward learning the average performance metrics intrinsic to each stock, marginally surpassing a simplistic constant price baseline. This suggests that, despite advanced architectures, relying solely on chart data places a ceiling on predictive capability—these networks appear to model "mean reversion" rather than genuine trend following. Consequently, the study underscores an essential limitation of historic price data as a solitary input source: the past is not necessarily a reliable oracle of future price trajectories in complex financial systems.

Radfar’s investigation further contextualizes this limitation by reflecting on the foundational assumptions of technical analysis—a field predicated on the discovery of recurring chart patterns to predict price movement. The findings cast significant doubt on the efficacy of these patterns, suggesting that many recognized signals may emerge as random occurrences rather than meaningful indicators. The apparent randomness reduces confidence in chart-based strategies and instead advocates for integrating multifaceted data sources capable of capturing underlying economic realities more effectively.

The study highlights the imperative role of fundamental analysis, emphasizing that a robust predictive model must synthesize diverse, high-dimensional inputs beyond raw price histories. Critical information streams such as financial statements, political developments, corporate product lifecycles, and broader economic indicators could be encoded into latent representations enriching the model’s contextual grasp. This blend of fundamental and technical features holds promise for transcending the simplistic paradigms of chart analysis and achieving more sophisticated stock trend inferences.

Intriguingly, Radfar remarks on the complexity and chaotic nature of financial markets—qualities that render them fertile testbeds for machine learning benchmarking. The intricacy of financial networks, their deeply entwined correlations across firms and sectors, and the persistent influence of exogenous shocks collectively challenge learning algorithms. Paradoxically, these characteristics, while obfuscating effective prediction, constitute a crucible for honing AI models’ generalizability and resilience.

The paper also distinguishes the operating dynamics of time series models from those of large language models (LLMs), underscoring that the former confront unique difficulties in handling noisy, non-stationary processes intrinsic to stock markets. Despite the recent surge in transformer-based LLMs, time series forecasting demands tailored architectures cognizant of its autoregressive and high volatility context. This reinforces the call for specialized network designs and training protocols attuned to financial temporal data’s idiosyncrasies.

One particularly salient insight revolves around data scale. Radfar’s experiments evince that models trained on limited stock market tickers—commonly the norm in financial machine learning datasets—simply lack the breadth to unearth robust predictive signals. Instead, predictive capability emerges only when models ingest datasets exponentially larger, involving hundreds or thousands of stocks across extensive time horizons. This suggests that sample diversity and volume are paramount, aligning with known “big data” principles but intensifying them in the financial realm.

Moreover, the paper raises critical attention to the evaluation metrics and validation methodologies underpinning financial forecasting research. It argues that research in this domain often overlooks the consequences of false positives and the reliability of positive signals in actual trading scenarios. This can lead to inflated performance perceptions and the adoption of models unfit for deployment—highlighting a pressing need for rigorous, real-world-oriented evaluation frameworks that mirror market complexities and operational constraints.

Radfar’s contribution is thus twofold: first, it filters out inflated claims regarding the predictive power of chart analysis and technical deep learning models; second, it lays the groundwork for more nuanced, integrative approaches marrying fundamental and technical data fusion. The ultimate goal is not merely to outsmart market noise but to construct models capable of navigating the multifactorial drivers influencing asset prices over time.

This study invites the financial AI community to rethink much of what is taken for granted in stock prediction paradigms. The seductive allure of pattern recognition on price charts is tempered with a sober acknowledgment that market behavior is influenced by a broader, interconnected ecosystem. Without incorporating multi-source data and expanding datasets’ scope dramatically, efforts at prediction may remain of limited utility.

In addition to methodological insights, Radfar’s work implicitly critiques the prevailing enthusiasm for “off-the-shelf” deep learning techniques in finance, suggesting that without domain-specific adaptations, these models falter when confronted with market realities. It encourages researchers to embrace interdisciplinary perspectives, weaving financial theory, econometrics, and machine learning into hybrid frameworks that better reflect economic fundamentals and stochastic market dynamics.

For practitioners, the implications are clear: reliance on technical indicators extracted from historical prices alone is insufficient. Successful deployment of algorithmic trading or portfolio management systems demands incorporating robust, external data, enhanced model validation, and considerable scale in training data. Only by navigating these complexities can AI-based financial forecasting approach genuine utility rather than mere academic curiosity.

Lastly, the study’s call for substantially larger datasets and more comprehensive input signals aligns with broader trends across AI research pushing towards data diversity and quantity as critical performance drivers. The stock market may well serve as a crucible for advancing time series forecasting methodologies on a global scale, with lessons extending beyond finance into other complex temporal domains.

Radfar’s revelations provide a reality check against overoptimism in neural network applications for financial trend prediction, highlighting both the challenges confronting the field and pathways forward through richer data integration and scaled experimentation. As stock markets continue to evolve amidst technological and geopolitical shifts, this research frames the cutting edge of AI’s potential and pitfalls in navigating one of the most baffling forecasting frontiers humanity confronts.

Subject of Research: Stock market trend prediction using deep neural networks and chart analysis

Article Title: Stock market trend prediction using deep neural network via chart analysis: a practical method or a myth?

Article References:
Radfar, E. Stock market trend prediction using deep neural network via chart analysis: a practical method or a myth?.
Humanit Soc Sci Commun 12, 662 (2025). https://doi.org/10.1057/s41599-025-04761-8

Image Credits: AI Generated