--- title: Econometrics & Forecasting url: https://www.ml-quant.com/topics/econometrics-forecasting/ site: ML-Quant (https://www.ml-quant.com) updated: 2026-10-09 license: Summaries CC BY 4.0; links go to the original sources index: https://www.ml-quant.com/llms.txt --- # Econometrics & Forecasting Forecasting, time series, econometrics and nowcasting. 315 papers featured; 36 in the last 12 months. Papers featured per quarter: 2023 Q2 8, 2023 Q3 33, 2023 Q4 21, 2024 Q1 30, 2024 Q2 34, 2024 Q3 36, 2024 Q4 32, 2025 Q1 33, 2025 Q2 40, 2025 Q3 12, 2025 Q4 7, 2026 Q1 3, 2026 Q2 0, 2026 Q3 8, 2026 Q4 18 ## Most cited - [TKAN: Temporal Kolmogorov-Arnold Networks](https://www.ml-quant.com/papers/ssrn/4825654/): 191 citations. The article presents Temporal Kolomogorov-Arnold Networks (TKANs), a new neural network design that merges the benefits of Recurrent Neural Networks and Long Short-Term Memory for improved multistep time series forecasting. - [Deep learning models for price forecasting of financial time series: A review of recent advancements: 2020–2022](https://www.ml-quant.com/papers/arxiv/2305.04811/): 176 citations. The article explores deep learning models for financial forecasting, discussing their structures, uses, pros and cons, and potential future research areas. - [Robust agents learn causal world models](https://www.ml-quant.com/papers/arxiv/2402.10877/): 107 citations. The research suggests that intelligent agents must learn an approximate causal model to generalize to new domains, impacting fields like transfer learning and causal inference. - [Universality of Linear Recurrences Followed by Non-linear Projections: Finite-Width Guarantees and Benefits of Complex Eigenvalues](https://www.ml-quant.com/papers/arxiv/2307.11888/): 43 citations. The study reveals that deep neural networks using linear complex-valued RNNs and MLPs can accurately approximate regular causal sequence-to-sequence maps, with complex eigenvalues near unit disk aiding in information storage. - [SST: Multi-Scale Hybrid Mamba-Transformer Experts for Time Series Forecasting](https://www.ml-quant.com/papers/arxiv/2404.14757/): 39 citations. The paper introduces the State Space Transformer model for time series forecasting, which effectively captures global and local patterns, offering superior performance with less memory and computational cost. - [Causal Diffusion Transformers for Generative Modeling](https://www.ml-quant.com/papers/arxiv/2412.12095/): 36 citations. The article discusses Causal Diffusion, a framework that enhances diffusion models' performance and allows a seamless shift between autoregressive and diffusion generation modes, achieving top results on the ImageNet generation benchmark. - [Assumption violations in causal discovery and the robustness of score matching](https://www.ml-quant.com/papers/arxiv/2310.13387/): 35 citations. The paper evaluates the performance of recent causal discovery methods on observational data, revealing that score matching-based methods excel in difficult scenarios, setting a new evaluation standard for causal discovery methods. - [Quantum-Enhanced Forecasting: Leveraging Quantum Gramian Angular Field and CNNs for Stock Return Predictions](https://www.ml-quant.com/papers/arxiv/2310.07427/): 32 citations. Quantum-Enhanced Forecasting for Time Series: The study introduces a time series forecasting method, Quantum Gramian Angular Field (QGAF), that merges quantum computing and deep learning to enhance the accuracy of time series classification and forecasting, and validates its effectiveness using major stock market datasets. - [Retail Demand Forecasting: A Comparative Study for Multivariate Time Series](https://www.ml-quant.com/papers/arxiv/2308.11939/): 29 citations. The study creates improved retail demand prediction models using macroeconomic factors and past sales data. - [Supervised autoencoder MLP for financial time series forecasting](https://www.ml-quant.com/papers/arxiv/2404.01866/): 28 citations. The study investigates the use of supervised autoencoders in improving financial forecasting through precise parameter tuning. - [The econometrics of happiness: Are We Underestimating the Returns to Education and Income?](https://www.ml-quant.com/papers/arxiv/1807.11835/): 27 citations. Value Rounding Behavior: The study addresses the issue of response scale simplification in surveys, particularly by less educated respondents, and introduces a model to estimate latent subjective wellbeing. - [Diffusive Gibbs Sampling](https://www.ml-quant.com/papers/arxiv/2402.03008/): 24 citations. The article introduces Diffusive Gibbs Sampling (DiGS), a new method for sampling from multi-modal distributions, which performs better in tasks like Bayesian neural networks and molecular dynamics. ## Latest - [StaFIR: Convex Learning of Stationarity-Aware Causal Filters](https://www.ml-quant.com/papers/arxiv/2610.07430/) (2026-10-09): A convex method learns causal filters that balance stationarity and input preservation for financial time series, adapting filtering strength to persistence while preserving signal similarity. - [Drivers of Exchange Rates in EMDEs: Implications for Foreign Exchange Intervention](https://www.ml-quant.com/papers/repec/imf-imfsdn-2026-003/) (2026-10-09): A framework using deviations from interest-rate parity identifies when exchange-rate movements stem from financial shocks rather than fundamentals, aiding intervention decisions. - [From Tone to Trajectory: Continuous Sentiment and the Shape of Monetary Policy Communication (Martin Feldkircher, Márton Kardos, Kristoffer Laigaard Nielbo)](https://www.ml-quant.com/papers/repec/onb-oenbwp-281/) (2026-10-09): The sequencing and emphasis of sentiment within central bank statements predict rate decisions and inflation expectations beyond average tone alone. - [Beyond the HAR: A New Benchmark for Volatility Forecasting](https://www.ml-quant.com/papers/repec/boa-wpaper-202648/) (2026-10-09): Fractional processes outperform discrete models like HAR for forecasting realized volatility at longer horizons by better balancing recent levels against long-run persistence. - [Macro News in Market Moves: Classifying News through Asset Co-movements](https://www.ml-quant.com/papers/repec/bca-bocsap-26-7/) (2026-10-09): CLONE decomposes daily price moves into demand, productivity, inflation, and policy news using stocks, bonds, and inflation swaps; aggregate demand dominated until 2021. - [Term Spread Volatility as a Leading Indicator of Economic Activity](https://www.ml-quant.com/papers/repec/hal-journl-hal-05740995/) (2026-10-09): Treasury yield curve slope volatility forecasts industrial production and employment growth at medium and long horizons, outperforming term spread alone after the 2008 crisis. - [Adaptive Bayesian Shrinkage of High-Dimensional Panel VARs](https://www.ml-quant.com/papers/repec/adl-wpaper-2026-05/) (2026-10-09): A Bayesian spike-and-slab method improves high-dimensional panel VAR forecasting and reveals heterogeneous spillover structures in sovereign bonds and macro data. - [Housing Network Connectedness and Policy Spillovers: Evidence from a Time-Varying Parameter VAR Approach](https://www.ml-quant.com/papers/repec/pre-wpaper-202630/) (2026-10-09): A time-varying VAR network reveals supply-inelastic coastal markets absorb largest regulatory shocks and spillovers persist to monetary conditions and inflation expectations. - [Perceived Monetary Policy Rules: Evidence from Professional Forecasters in Australia](https://www.ml-quant.com/papers/repec/adl-wpaper-2026-02/) (2026-10-09): Time-varying beliefs about the central bank's reaction function collapsed at the zero bound and rose only after observed tightening, not from guidance alone. - [Do Your Own Research: Learning to Forecast by Learning to Search](https://www.ml-quant.com/papers/arxiv/2610.01955/) (2026-10-02): Training teaches a language model to improve event forecasting by learning which evidence to gather, beating frontier models on hard questions at lower cost. - [DualCast: A Dual-Path Language Model for Bimodal Financial Time-Series Forecasting](https://www.ml-quant.com/papers/arxiv/2609.38197/) (2026-10-02): Dual-path language model extended with financial tokens and news-conditioned revision achieves lowest error in 8 of 12 equity and energy forecasting settings across multiple horizons. - [Market History, Rebuilt A Reproducible Synthetic U.S. Total-Return Dataset for Leverage and Duration Research ChemicalStats • Anonymous Risk Researcher](https://www.ml-quant.com/papers/ssrn/7527762/) (2026-10-02): A new 234-year synthetic total-return dataset for equities and Treasuries reveals that permanent 3× leverage faces near-total loss in most reconstructions, ranking below 2× on risk-adjusted returns. - [Bayesian Vector Autoregression for Joint Macroeconomic Forecasting: Pseudo-Real-Time Evidence, Density Forecasts, Structural Dynamics, Scenario Analysis, and Practical Macro-Risk Use for the United States](https://www.ml-quant.com/papers/ssrn/7532698/) (2026-10-02): The research develops a compact Bayesian VAR for real-time forecasting of GDP growth, inflation, unemployment, and the federal funds rate, with evaluation emphasizing density scores and predictive-interval diagnostics. - [The macroeconomic effects of AI technology shocks](https://www.ml-quant.com/papers/repec/bdi-wptemi-td-1542-26/) (2026-10-02): AI-intensive patents generate delayed productivity and employment gains alongside falling consumer prices, with substantially larger aggregate effects than broader ICT shocks but reducing labor share and increasing wealth inequality. - [The Uncertainty Channel of Monetary Policy Communication](https://www.ml-quant.com/papers/repec/tse-wpaper-132139/) (2026-10-02): The study shows that increased Federal Reserve communication lowers monetary policy uncertainty and generates substantial real effects: industrial production rises 0.3 percent and unemployment falls 0.2 percentage points within two months. - [What makes Monetary Policy More Powerful? A Big Data Approach](https://www.ml-quant.com/papers/repec/cbi-wpaper-09-rt-26/) (2026-10-02): Analysis of a large macro-financial dataset ranks non-linear monetary transmission channels, finding transmission to long-term rates weakens at high interest rates and high credit growth, with sovereign risk mattering in the euro area. - [Price Conflict and US Stock Return Volatility Forecasting: Insights from over 150 Years with a Mixed-Frequency Framework](https://www.ml-quant.com/papers/repec/pre-wpaper-202620/) (2026-10-02): The GARCH-MIDAS model incorporating a quarterly news-based Price Conflict Index outperforms benchmarks for forecasting US stock volatility over 150 years of monthly and daily data. - [Zero-Shot Conditional Forecasting and the Information Content of Central Bank Paths](https://www.ml-quant.com/papers/repec/ces-ceswps-12972/) (2026-10-02): A pre-trained time-series model reading central bank published paths cuts forecast errors better than the banks themselves and hard-conditioned VARs, revealing exploitable institutional differences. - [The Impossible Trinity of Time-Series Validation: A Conservation Law among Training Sufficiency, Test Coverage, and Temporal Causality](https://www.ml-quant.com/papers/arxiv/2609.29530/) (2026-09-25): Proves that training sufficiency, test coverage, and temporal causality cannot be maximized simultaneously in time-series validation, pricing each constraint explicitly. - [Hierarchical Multi-Task Learning with Liquidity-Aware Signals for Stock Forecasting](https://www.ml-quant.com/papers/arxiv/2609.25617/) (2026-09-25): A hierarchical multi-task framework jointly predicts price movement, volatility and volume using liquidity-aware signals, outperforming neural and tree-based baselines on Chinese equity indices. - [Stealing profits: Spread-based temporal hierarchy forecasting for day-ahead electricity markets](https://www.ml-quant.com/papers/arxiv/2609.23223/) (2026-09-25): Jointly reconciling hourly price and spread forecasts improves intraday electricity price prediction accuracy by up to 19.7% and battery-arbitrage profits by up to 10.4%. - [Nonlinear Drivers of Macroeconomic Tail Risk: A Threshold Stochastic Volatility-in-Mean VAR with Regime-Dependent Leverage](https://www.ml-quant.com/papers/arxiv/2609.26994/) (2026-09-25): A regime-switching volatility-in-mean VAR reveals that the drivers of growth and inflation tails differ from median dynamics, with macroeconomic uncertainty playing a larger role in downside risk. - [A Stochastic Nested Fixed Point Algorithm for Large-Scale BLP Estimation](https://www.ml-quant.com/papers/arxiv/2609.23998/) (2026-09-25): A stochastic nested fixed-point estimator reduces memory and computational cost for random-coefficients logit demand models, enabling estimation on 100 million markets in hours. - [The "Rough" HAR Model](https://www.ml-quant.com/papers/arxiv/2609.21587/) (2026-09-25): Augmenting HAR with a negative moving-average component approximates rough dynamics, outperforming classical models out-of-sample and matching continuous-time rough model accuracy. - [Network Realized GARCH--Itô Models: Volatility Spillovers with High-Frequency Identification](https://www.ml-quant.com/papers/arxiv/2609.29515/) (2026-09-25): Introduces a network realized GARCH-Itô model that identifies dynamic volatility transmission among assets using high-frequency data, outperforming recursive forecasts on sector ETFs. - [Industry Information and Equity Return Predictability](https://www.ml-quant.com/papers/ssrn/7486138/) (2026-09-25): Using production, employment, and sales data across 426 industries, the research shows that upstream industry signals predict aggregate monthly stock returns with 23.8% out-of-sample R-squared. - [Discounted Sales of Expiring Perishables: Challenges for Demand Forecasting in Grocery Retail Practice](https://www.ml-quant.com/papers/arxiv/2602.04464/) (2026-02-12): Including discounted sales of soon-to-expire perishables in demand forecasts leads to underestimating demand, highlighting the need for better forecasting to reduce inventory waste in grocery stores. - [Directional-shift Dirichlet ARMA models for compositional time series with structural break intervention](https://www.ml-quant.com/papers/arxiv/2601.16821/) (2026-02-02): The article introduces a new Bayesian model that analyzes compositional time series data, effectively handling structural breaks and enhancing forecasting accuracy during these changes. - [History Is Not Enough: An Adaptive Dataflow System for Financial Time-Series Synthesis](https://www.ml-quant.com/papers/arxiv/2601.10143/) (2026-01-16): A new adaptive data management system enhances model performance in quantitative finance by constantly updating to reflect market changes, addressing the shortcomings of relying solely on historical data. - [Sample Size Issues in Finance Research](https://www.ml-quant.com/papers/ssrn/4463148/) (2025-12-28): The study promotes the use of Bayesian statistics in finance to better analyze large AI-generated datasets and mitigate misleading significance from traditional methods.