arXiv:2608. 12259v1 Announce Type: new Abstract: Financial forecasting models are typically developed in full precision, yet production deployment often requires low-precision inference to reduce memory and computational cost.
arXiv:2510. 04487v5 Announce Type: replace Abstract: While accuracy is a critical requirement for time series forecasting, an equally important desideratum is reasonable forecast volatility across forecast creation dates (FCDs).
arXiv:2606. 27688v1 Announce Type: cross Abstract: In financial forecasting, predictive performance depends not only on which model is trained, but also on how the trained model is deployed.
The paper introduces PRICE, a systematic framework for adapting Large Language Models to short‑term Bitcoin price forecasting. PRICE combines parameter‑efficient fine‑tuning with LoRA, recursive multi‑step inference, integer‑rounded numerical representation, Context‑Task‑Format prompting, and exact zero‑temperature decoding, all built on a 4‑bit quantized LLaMA‑3 8B model. Ablation studies and comparative evaluations show that each component improves accuracy and reliability, enabling PRICE to achieve the lowest forecasting errors among eight transformer‑based and time‑series foundation models.
arXiv:2607. 16229v1 Announce Type: cross Abstract: Large language models (LLMs) are increasingly used as components of agentic systems that observe, plan, and act.
arXiv:2608. 27076v1 Announce Type: new Abstract: Algorithmic trading now represents a market exceeding $20 billion, where even marginal gains in signal robustness can translate into economically significant returns.
arXiv:2606. 03184v1 Announce Type: cross Abstract: Financial forecasting is difficult due to low signal-to-noise ratios, latent factors, heavy tails, regime shifts, and jumps.
arXiv:2606. 04980v1 Announce Type: new Abstract: Mixture-of-Experts (MoE) architectures scale model capacity through sparse expert activation, but their deployment remains memory-bound because all expert weights must reside in memory.
UQ-LOB is a lightweight, encoder‑agnostic module that adds uncertainty quantification to any pretrained limit order book (LOB) encoder. It offers two variants: UQ‑regression, which outputs a calibrated Gaussian over future tick displacement, and UQ‑classification, which outputs a categorical distribution over down/up/stationary. On 5.2 billion LOB events across seven cryptocurrency assets, UQ‑regression achieves near‑nominal 68 % interval coverage, and selecting the top 10 % most confident predictions boosts directional macro F1 by 0.11–0.15 for regression and 0.05–0.11 for classification, reaching F1 scores of 0.88 (down) and 0.83 (up) at a 5‑second horizon.
DualCast is a dual‑path language model that forecasts financial time‑series by combining a fast numerical forecaster with an optional text‑conditioned revision mechanism. The fast path trains only new financial‑token embeddings and output heads on a frozen Qwen3‑8B backbone, while the slow path uses a LoRA adapter to incorporate news and refine predictions. In zero‑shot tests across equities and energy prices at multiple time resolutions, the slow path achieves the lowest mean absolute percentage error in most settings, especially for longer horizons, and news ablations show additional gains in many markets.
arXiv:2602. 15327v2 Announce Type: replace-cross Abstract: Machine learning model performance improvements tend to arise from competition and application.
VertiFuseX is a hybrid LSTM architecture that fuses multi‑scale temporal representations at the penultimate layer, stacking features from LSTM, Bi‑LSTM, and St‑LSTM branches and a parallel DNN stream. On 15 years of global equity index data, it reduces MAPE by 30‑54% and improves MAE and RMSE by over 40% compared to LSTM baselines, outperforming seven state‑of‑the‑art models across 33 metric‑dataset comparisons. The model is lightweight (675k parameters, 2.6 MB footprint) with 1.5 ms/sample inference latency and demonstrates robust, interpretable forecasting with reduced drawdowns in algorithmic trading simulations.
Aurora‑X is a billion‑parameter time‑series foundation model designed for extreme forecasting tasks. It employs a progressive curriculum that starts with channel‑independent pretraining, then adds cross‑variable dependencies, variable context and horizon lengths, and optional future covariates during mid‑training. A variable‑resolution post‑training stage allows adjustable temporal spans per token at inference, while a pattern‑guided mixture‑of‑experts expands capacity through sparse activation and expert specialization. An implicit quantile network head predicts arbitrary quantiles, enhancing probabilistic forecasting flexibility. Experiments on GIFT‑Eval, TIME, FEV‑Bench, TFB, and DAG‑Bench show state‑of‑the‑art performance against both pretrained TSFMs and task‑specific supervised models.