|
on Forecasting |
| By: | Aitor Ciarreta; Peru Muniain; Ainhoa Zarraga |
| Abstract: | The increasing volatility of electricity prices driven by renewable energy integration, market shocks, and regulatory changes has reinforced the need for forecasting methods that go beyond point predictions and accurately describe the full conditional price distribution. This paper applies the Generalised Additive Models for Location, Scale and Shape (GAMLSS) framework to forecast Spanish day-ahead electricity prices using hourly data from 2020 to 2024. Alternative specifications based on Normal, Johnson's SU (JSU), and Sinh-Arcsinh (SHASH) distributions are considered, allowing the location, scale, and shape parameters to vary with market fundamentals, including electricity demand, renewable generation, seasonal effects, and regulatory and geopolitical risk factors. Forecasts are generated using a rolling-window approach and evaluated through the mean absolute error (MAE), pinball loss, and Diebold-Mariano tests. The results show that flexible distributional specifications improve forecasting performance relative to a naive benchmark and the standard normal specification. While SHASH and JSU specifications provide the lowest point forecasting errors, the hourly analysis reveals substantial intraday variation in relative performance across specifications. JSU specification with all four parameters driven by covariates achieves the best probabilistic forecasting performance, particularly in the tails of the distribution. Diebold-Mariano tests confirm the statistical significance of these improvements. These findings highlight the importance of modelling time-varying shape distributional parameters and demonstrate the value of GAMLSS models for forecasting and risk management in increasingly volatile electricity markets. |
| Date: | 2026–10 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.01465 |
| By: | Hengyi Yang; Sida Lin; Yiyan Qi; Yankai Chen; Haohan Zhang; Xianhua Peng; Jian Guo |
| Abstract: | Stock price forecasting is a long-standing challenge in computational finance, driven by the inherent randomness of markets and complex temporal patterns. While recent deep-learning models have raised forecasting accuracy by jointly modeling inter-stock and temporal price dynamics, they conflate inter-stock relationships with intra-stock temporal dependencies and focus solely on the univariate objective of price movement. To address these limitations, we propose LiMT, a Hierarchical Multi-Task Learning framework that integrates liquidity-aware signals for stock price forecasting. LiMT employs a Market Regime Encoder (MRE) module that first extracts contemporaneous cross-stock dependencies, then models each stock's temporal dynamics, yielding a unified latent state. Building on this latent state, we introduce a Liquidity-Driven Learning (LDL) module, a mixture-of-experts architecture that features cross-task gating mechanisms to jointly predict price movement, volatility, and trading volume. We further design an Adaptive Portfolio Optimization (APO) mechanism that converts multi-task forecasts into executable portfolio weights under transaction-cost and liquidity constraints. Extensive experiments on the CSI300 and CSI500 benchmarks show that LiMT performs best among strong neural and tree-based baselines across the reported metrics. In realistic CSI300 backtests, APO improves annualized return from 3.99% to 10.01% and Sharpe ratio from 1.22 to 1.86 over equal weighting, showing that the multi-task forecasts translate into deployable portfolio gains. |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2609.25617 |
| By: | Tamkin Nuriyev (Central Bank of Azerbaijan Republic); Aygun Garayeva (Central Bank of Azerbaijan Republic); Gulzar Tahirova (Central Bank of Azerbaijan Republic) |
| Abstract: | Using 800, 000 transaction-level customs records from January 2018 to February 2026, the paper constructs a trade-weighted Imported Food Price Index (IFPI), covering 34 items from the consumer basket with significant import dependence. The index is developed using the Fisher ideal methodology to provide a timely measure of external food price pressures. The results indicate that the IFPI leads official food Consumer Price Index (CPI) by approximately two months, with a maximum correlation of 0.81, highlighting its potential usefulness as an early indicator of domestic food inflation. Building on this, the paper develops a forecasting framework for the IFPI by combining non-parametric Binary Segmentation and Hidden Markov Models with a regularized machine-learning ensemble. The model employs an ensemble approach that combines Histogram-based Gradient Boosting Regression Tree, Random Forest, and Extreme Gradient Boosting, alongside rigorous time-series cross-validation. The optimized ensemble achieves a 58% out-of-sample R² relative to a random walk benchmark, vastly outperforming traditional linear Autoregressive Distributed Lag (ARDL) (13.60%) and Autoregressive Integrated Moving Average with Exogenous Variables (ARIMAX) (0.18%) baselines. The forecast results are intended to be incorporated into broader inflation forecasting models to improve short-term projections. |
| Keywords: | Import price index; Fisher Ideal index; Food price inflation; Machine learning forecasting; Hidden Markov models. |
| JEL: | C43 C53 C55 E31 F14 |
| Date: | 2026–01–16 |
| URL: | https://d.repec.org/n?u=RePEc:aze:wpaper:2601 |
| By: | Zhiruo Zhang (Adelaide University); Firmin Doko Tchatoka (Adelaide University); Qazi Haque (Adelaide University) |
| Abstract: | We develop a Bayesian framework that combines adaptive shrinkage with variable selection to address over-parameterisation and sparsity in high-dimensional panel vector autoregressions (PVARs). The proposed approach employs Laplace-based spike-and-slab priors to enable flexible modelling of dynamic cross-sectional interdependencies and unit-specific heterogeneity. Monte Carlo evidence shows that the method delivers improvements in estimation accuracy and forecasting performance relative to existing regularisation approaches. We illustrate its empirical relevance in two applications. The first investigates financial contagion in euro area sovereign bond markets, while the second examines international forecasting performance in a multi-country macroeconomic panel. The results highlight the benefits of adaptive, component-specific shrinkage for capturing heterogeneous spillover structures in complex panel systems. |
| Keywords: | Dynamic Interdependency;Cross-sectional Heterogeneity;Bayesian Lasso;Variable Selection;Spike and Slab Prior;Financial Contagion |
| Date: | 2026–05 |
| URL: | https://d.repec.org/n?u=RePEc:adl:wpaper:2026-05 |
| By: | Masoud Soleimani |
| Abstract: | Forecasters often score the same units per date against one standardized realized outcome. We show that every standardized forecast splits exactly into a component aligned with this common target and a component uncorrelated with it. Three consequences follow: forecast-error correlation largely mirrors forecast correlation and is therefore a poor measure of diversity; an equally weighted combination beats a no-information forecast only when average alignment is large relative to the combination's dispersion; and the gain from adding a forecaster separates into genuine improvement and mere dilution, which equal-weight admission can mistakenly reward. We develop a cautious selection rule, study it in simulations, and apply it to language-model forecasts of US equity rankings and mechanical signals ranking exchange-traded funds. Selection removes most dilution losses, but no combination beats the no-information forecast. |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2609.26303 |
| By: | Andrzej Tokajuk; Jaros{\l}aw A. Chudziak |
| Abstract: | Volatility forecasts play a central role in financial risk management because their overall level and day-to-day movements affect downstream decisions. Most studies compare forecasting models while keeping the training loss fixed. Yet losses emphasise different errors and can target different properties of future volatility, so raw comparisons may combine persistent forecast-level differences with differences in daily forecast movements. This leaves unresolved whether the importance of loss choice comes mainly from the forecast level it targets or from differences that remain after level adjustment. We address this gap through a comparison of seven losses and five models across major cryptocurrencies. Validation-based alignment adjusts the forecast level before the raw and aligned forecasts are evaluated using statistical scores and one-day Value-at-Risk. Before alignment, marginal score variation is greater across losses. After alignment, model choice becomes the larger source of variation in the full five-model comparison, while cross-loss differences in VaR breach rates narrow substantially. Our contribution is a comprehensive evaluation of loss and model choice that shows why losses can appear so influential in raw comparisons and how this interpretation changes when forecast level and downstream risk are considered explicitly. |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2609.27024 |
| By: | Nabeel Ahmad Saidd |
| Abstract: | Financial time series evolve across multiple temporal resolutions, challenging forecasting systems to incorporate newly available information without repeatedly recomputing unchanged representations. We introduce HARN, a Hierarchical Associative Resonance Network for event-driven multi-timeframe forecasting. HARN maintains persistent representations across temporal levels and updates each level only when its corresponding completed bar becomes available. The architecture combines causal multi-scale temporal encoding, gated associative memory, cross-level resonance, and hierarchical evidence aggregation, with forecasting performed in basis-point space and reconstructed to the original price scale. We evaluate HARN on four assets spanning equity, foreign exchange, and commodity markets using multiple random seeds and component ablations. HARN achieves competitive reconstructed-price forecasting errors against single-timeframe PatchTST and TimeXer baselines, while ablations reveal the effects of removing individual components across assets and timeframes. A code-level audit further examines consistency between the implementation and the defined event-driven causal protocol. The results position HARN as a persistent multi-timeframe forecasting framework rather than evidence of universal predictive superiority. |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2609.26822 |
| By: | Michela Eugenia Pasetto (Bank of Italy) |
| Abstract: | The accurate valuation of unquoted shares on the liability side of non-financial corporations' balance sheets is of primary importance, as equity issuance often constitutes a strategic choice for raising capital to finance investments. At macroeconomic level, quarterly data on equity issuance are published in financial accounts statistics and rely on estimation techniques. This paper proposes a new estimation method based on a Bayesian vector autoregression model to forecast unquoted equities issued by Italian non-financial corporations on a quarterly basis. The model is applied to both outstanding amounts and transactions. We compare the performance of the proposed model in its point forecasts and variability with five alternative econometric approaches, showing it performs very well in out-of-sample forecasting. |
| Keywords: | Bayesian VAR, forecasts, financial accounts, equity |
| JEL: | C32 C53 E27 E66 |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:bdi:opques:qef_1037_26 |
| By: | Schröder, Maximilian |
| Abstract: | Macroeconomic policy increasingly requires assessments of risk rather than point forecasts alone. This paper develops a framework to compute impulse responses for scenario probabilities and the full predictive distribution of macroeconomic outcomes. The approach combines a regime-dependent mixture VAR with endogenous state probabilities and heteroskedastic shock identification, complemented by sparse sign and narrative restrictions. Applied to the euro area and U.S., the framework addresses a central policy question following the outbreak of the 2026 Iran war: do geopolitical shocks imply a return to the risk profile observed after Russia’s invasion of Ukraine, or is this time different? Results show that shocks differ not only in their effects on expected outcomes, but also in how they reshape scenario probabilities, tail risks, and transitions into high-volatility states. JEL Classification: C11, C32, C34, C53, E37 |
| Keywords: | Bayesian econometrics, density forecasting, macroeconomic risks, scenario analysis, state-dependent transmission |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:ecb:ecbwps:20263290 |
| By: | Farid Muradzada (Central Bank of Azerbaijan Republic); Khazan Bakhshaliyev (Central Bank of Azerbaijan Republic) |
| Abstract: | This study develops a forecasting framework for Brent crude oil prices based on the fundamental drivers of the global oil market, namely demand, supply, and market expectations. The estimation is conducted using quarterly time-series data covering the period 1994–2025 and employs a Vector Autoregressive (VAR) model. The results indicate that the proposed model demonstrates strong forecast performance under conditions of high oil price volatility. The forecasting error of the model is lower than that of benchmark models. Furthermore, the findings confirm that geopolitical tensions, regional conflicts, and uncertainty in global energy markets have a significant impact on oil price dynamics through oil futures prices. |
| Keywords: | Brent crude oil price, oil price forecasting, VAR model |
| JEL: | C53 Q43 Q47 |
| Date: | 2025–08–05 |
| URL: | https://d.repec.org/n?u=RePEc:aze:wpaper:2503 |
| By: | Paul Hubert; Rose Portier |
| Abstract: | Are individuals who are most confident in their forecasts the most accurate? This is what household surveys reveal, unlike professional forecasters. Drawing on four surveys from the euro area and the United States, we show that the uncertainty reported by households regarding their inflation forecasts contains information about the quality of those forecasts, particularly when households overestimate inflation. <p> Les individus les plus confiants dans leurs prévisions sont-ils les plus précis ? C’est ce que révèlent les enquêtes auprès des ménages, contrairement aux prévisionnistes professionnels. À partir de quatre enquêtes en zone euro et aux États-Unis, nous montrons que l'incertitude déclarée par les ménages sur leurs prévisions d’inflation contient une information réelle sur la qualité de ces prévisions, et ce d’autant plus lorsque les ménages surestiment l’inflation. |
| Date: | 2026–10–02 |
| URL: | https://d.repec.org/n?u=RePEc:bfr:econot:461 |
| By: | Guido Bulligan (Bank of Italy); Paolo Farroni (Bank of Italy) |
| Abstract: | We examine the post-COVID relationship between household-level income uncertainty and consumption-saving behaviour, using granular data from the ECB Consumer Expectations Survey (CES). We find that higher uncertainty is associated with higher saving rates and lower nominal consumption, with stronger responses among lower-income and liquidity-constrained households. By decomposing uncertainty into downside and upside components, we show that most of the effect is driven by downside risk. At country level, aggregated income uncertainty provides some information beyond the traditional determinants of national saving rates, although its contribution to improving macro-model point forecast accuracy is limited. |
| Keywords: | uncertainty, income risk, saving rate, household expectations, liquidity constraints, macroeconomic forecasting |
| JEL: | D12 D84 E21 D91 |
| Date: | 2026–07 |
| URL: | https://d.repec.org/n?u=RePEc:bdi:opques:qef_1046_26 |
| By: | Lucas Harlaar (Maastricht University); Jan van den Brakel (Maastricht University); Siem Jan Koopman (Vrije Universiteit Amsterdam) |
| Abstract: | We develop a multivariate time series model for the complete national accounts dataset which typically consists of at least nine variables. The dynamic features of the variables are characterized by a set of unobserved components that evolve stochastically over time and may represent trend, cyclical, and seasonal effects. The components can be uniquely associated with a single variable or can be common to multiple variablesThe model treats expenditure and production variables jointly, subject to accounting-identity constraints, ensuring that their gross domestic product aggregates are consistent in both mean and variance by construction. The model constraints can be treated naturally within the Kalman filter. Parameter estimation is based on exact maximum likelihood which is feasible despite the high-dimensional parameter vector. This model-based framework allows for synchronized nowcasting, forecasting, seasonal adjustment, trend and cycle extraction for national accounts variables, including gross domestic product. We show that the statistical precision of signal extraction and forecasting improves when our framework is used instead of univariate and unconstrained model alternatives. When the model includes a common cycle component for all national accounts variables, it can be interpreted as a macroeconomic cycle indicator at current prices. We empirically illustrate our model-based methodology for data from national accounts of Germany, Italy, and the Netherlands, and show its significance for both official statistics and macroeconomic policy-making. |
| JEL: | C32 C54 |
| Date: | 2026–09–27 |
| URL: | https://d.repec.org/n?u=RePEc:tin:wpaper:20260073 |
| By: | Aojie Yuan; Haiyue Zhang; Zijian Su |
| Abstract: | Decision-focused learning (DFL) trains predictors through downstream objectives, but a different loss need not provide an independent parameter-update direction. We characterize this restriction through the predictor Jacobian, using sparse index tracking to distinguish the covariance entries read by the optimizer from the parameter directions available to learning. Rank-one Jacobians make nonzero per-example gradients collinear; a conditional spectral bound describes near-collinearity. A batch-subspace characterization and counterexamples show why these local statements imply neither common minimizers nor collinear batch updates. Experiments examine when geometry translates into decision quality. Across 38 one-parameter equity configurations, DFL gains over MSE remain below 1.8%; a 385-parameter conditional predictor also has pointwise rank one. In validation-tuned shortest-path and knapsack experiments, full-capacity SPO+ reduces mean regret by 11.6% and 10.6%, respectively; only knapsack survives correction across eight comparisons. The capacity contrast persists on fresh datasets across batch orders and training budgets. Holding expressivity fixed, invertible coordinate scaling lowers spectral effective rank and ordinary SGD gains; compensating for the scaling restores the original trajectories. Financial forward-target controls separate forecast accuracy from decision quality; a matched neural comparison finds no aggregate DFL advantage in the tested architecture. These findings distinguish local rank restrictions, coordinate-dependent optimization and predictive accuracy. Predictor geometry helps explain available learning directions, while held-out decision quality remains the test of practical benefit. |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2609.39261 |
| By: | Pierre Aldama; Alice Carroy; Bruno Ducoudré; Raphaël Martin; Mylène Sabatini; David Sabes |
| Abstract: | This paper has two objectives. First, it updates the set of Basic Model Elasticities (BMEs) of the Banque de France's semi-structural macroeconomic model for France (FR-BDF), initially presented in 2020. Since then, FR-BDF has been enhanced with detailed blocks for household credit and housing markets, corporate financing, and fully re-estimated following the release of national accounts in base 2020. The BMEs reported here reflect these improvements and remain constructed under an orthogonality principle, providing analytical responses to isolated shocks that can be combined to design alternative scenarios. Second, the paper introduces FR-BMEs, a simplified, linearized model based on FR-BDF BMEs, which has been used since March 2024 for interim projections. While this approach improves efficiency, it reduces the granularity of projections, notably for public finances and household income. FR-BMEs also supports policy analysis, including model-based forecast-error decomposition. We illustrate this with a post-mortem of the 2024 GDP growth forecast error from the December 2023 Banque de France's projection. Actual growth reached 1.1 % versus a forecast of 0.9 %, implying a +0.2pp error. This discrepancy mainly reflects historical data revisions (+0.16pp) and, to a lesser extent, changes in Eurosystem assumptions (+0.05pp). By coincidence, revisions in public finance assumptions (+0.28 pp) and errors in expert judgement (- 0.28pp) offset each other. |
| Keywords: | Semi-Structural Modelling, Macroeconomic Forecasting, Macroeconomic Policy Analysis |
| JEL: | E17 E6 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:bfr:banfra:1045 |
| By: | Xinyu Song |
| Abstract: | We introduce a network realized GARCH-It\^o model in which volatility transmission is a dynamic relation among the latent daily integrated volatilities of multiple assets. An unknown directed and signed network is embedded in a continuous-time variance process and appears in the resulting exponential daily recursion. Intraday returns identify integrated volatility and hence the network that governs its propagation. For a fixed network template, feasible estimation based on realized volatility is first-order equivalent to estimation based on latent integrated volatility. For an unknown network, regularization selects the relevant structure, and, under oracle conditions, a fixed-rank local refit provides conditional inference for model-implied response-based connectedness. In an application to nine U.S. sector ETFs from 2007 to 2025, the model attains the lowest average out-of-sample QLIKE among the reported recursive forecasts, although its advantage over HAR is not statistically significant. In the full sample the weighted LASSO selects an empty sparse support, so inference concerns the rank-one projected factor component; this conditional analysis identifies Energy as a net volatility transmitter. Aggregate connectedness is more stable across volatility measures than individual sparse channels. |
| Date: | 2026–08 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2609.29515 |
| By: | Edward P. Herbst; Karen Page |
| Abstract: | Multiple-horizon forecast panels are increasingly used to infer perceived monetary policy rules, but inference depends on how coefficients are pooled across forecasters, dates, and horizons. We treat this pooling structure as the object of inference. In a participant-date-horizon Taylor-rule regression model, we compare pooling patterns using Bayesian marginal likelihoods, applying the framework to the Blue Chip Financial Forecasts, Survey of Professional Forecasters, and the Summary of Economic Projections. The preferred specifications place much of the systematic variation in policy-rate forecasts in intercepts that vary across forecast horizons and survey dates. Evidence of "changing perceptions" of monetary policy via economically meaningful time-varying response coefficients is weak overall. |
| Keywords: | Taylor rules; monetary policy expectations; survey forecasts; Bayesian model selection; coefficient heterogeneity |
| JEL: | C11 C23 E47 E52 |
| Date: | 2026–09–18 |
| URL: | https://d.repec.org/n?u=RePEc:fip:fedgfe:103790 |
| By: | Zeno Enders; Sophia Oberbrinkmann |
| Abstract: | Many central banks, such as the ECB, explicitly tie monetary policy decisions to medium-term inflation forecasts. Standard New Keynesian models, however, typically abstract from this feature. This paper studies equilibrium determinacy in a New Keynesian model in which monetary policy targets inflation at a forecast horizon larger than one. When policy reacts to sufficiently distant inflation expectations, it may fail to pin down a unique equilibrium. In fact, stronger reactions to expected inflation can make indeterminacy more likely. Whether monetary policy can achieve determinacy at a given forecast horizon depends not only on satisfying the Taylor principle, which provides a lower bound on the inflation coefficient, but also on an upper bound. This upper bound becomes more restrictive as the forecast horizon increases, until determinacy becomes impossible, and depends critically on the slope of the New Keynesian Phillips curve. A sufficiently flat Phillips curve preserves a non-empty determinacy region even at longer forecast horizons. Introducing fiscal policy can restore determinacy across all forecast-horizon targets by ensuring a unique stationary equilibrium. However, this holds only if fiscal policy does not actively stabilize government debt, which implies that inflation forecast targeting at medium horizons puts monetary policy in a passive position and invites fiscal dominance. |
| Keywords: | inflation forecast targeting, fiscal policy, monetary–fiscal interaction, determinacy, New Keynesian models, forward-looking Taylor rules |
| JEL: | E52 E58 E63 |
| Date: | 2026 |
| URL: | https://d.repec.org/n?u=RePEc:ces:ceswps:_12991 |
| By: | Shuping Shi (Macquarie University); Jun Yu (University of Macau); Chen Zhang (Sun Yat-sen University) |
| Abstract: | Continuous time models provide a stronger benchmark for realized volatility forecasting than the discrete time models standard in empirical finance. Using 111 U.S. equities and ETFs, we compare continuous time fractional processes, ARFIMA, and HAR models on a common footing. Three families perform similarly at the daily horizon but separate at longer horizons, where continuous time models lead statistically and economically. The ranking reflects how each model balances recent volatility against its long run level. The continuous time advantage grows with volatility persistence and diminishes as jumps account for more variation. |
| Keywords: | Realized volatility, Continuous time models, Discrete time models, Forecasting, Economic utility |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:boa:wpaper:202648 |
| By: | Colin Aitken; Michael K. Tippett; Pedram Hassanzadeh; Katherine Kowal; Rendani Mbuvha; John H. Marsham; Shruti Nath; Ousmane Ndiaye; Douglas J. Parker; Caroline M Wainwright; Michael Kremer; William R. Boos |
| Abstract: | Artificial-intelligence weather prediction (AIWP) models have made it possible to produce high-quality tailored forecasts with limited computational resources. This advance has the potential to benefit hundreds of millions of farmers in low- and middle-income countries who lack access to forecasts of critical weather phenomena. However, it can be difficult for key stakeholders to evaluate forecast quality, risking a "race to the bottom" as cheap but low-quality forecasts crowd out forecasts that would benefit farmers. We propose a set of principles and protocols for evaluating agriculturally-relevant forecasts as a starting point for standards that would let forecasters credibly convey their forecasts' quality. |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2610.00782 |
| By: | Joel Pfeffer (Allora Foundation); J. M. Diederik Kruijssen (Allora Foundation); Florian Stecker (Allora Foundation); Steven N. Longmore (Allora Foundation; LJMU) |
| Abstract: | In quantitative finance, standard regression losses are misaligned with the economics of return prediction. As the conditional mean of financial log-returns is close to zero, symmetric losses such as the mean squared and mean absolute errors make the constant zero forecast a near-optimal solution, penalizing models with genuine but noisy directional skill. This applies both during training, where predictions shrink toward zero, and during evaluation, where trivial forecasters can lead loss-based rankings. Under a Gaussian linear prediction model, we show that all symmetric monotonic losses share a universal breakeven directional accuracy against the zero predictor, which rises sharply and becomes unobtainable as the prediction noise approaches the standard deviation of the returns. We introduce the CZAR (Composite Zero-Agnostic Return) loss function, a piecewise quadratic loss built around five key requirements: convexity in the prediction, asymmetry with true return direction that vanishes at zero, near-linear penalization of undershoots and wrong-direction predictions, divergence for large errors, and an adaptive loss floor for evaluation. CZAR is provably convex in the prediction at fixed true value, has closed-form gradient and Hessian suitable for custom objectives in gradient-boosted libraries, and its four hyperparameters reduce to a single choice through correlated defaults. In idealized tests, the minimum directional accuracy required for a CZAR-evaluated forecaster to outperform the zero predictor under mean log loss remains near the 50% chance level, whereas the corresponding threshold for symmetric losses rises sharply with prediction noise. This advantage persists under heavy-tailed return distributions. In a LightGBM experiment on intraday BTC log-returns, CZAR-trained models reduce the `zero-returns bias' and improve directional accuracy on large-magnitude returns. |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:arx:papers:2609.36061 |
| By: | Zongwu Cai (Department of Economics, The University of Kansas, Lawrence, KS 66045, USA); Ying Fang (The Wang Yanan Institute for Studies in Economics, Xiamen University, Xiamen, Fujian 361005, China and Department of Statistics & Data Science, School of Economics, Xiamen University, Xiamen, Fujian 361005, China); Ming Lin (The Wang Yanan Institute for Studies in Economics, Xiamen University, Xiamen, Fujian 361005, China and Department of Statistics & Data Science, School of Economics, Xiamen University, Xiamen, Fujian 361005, China); Zeqin Liu (School of Statistics, Shanxi University of Finance and Economics, Taiyuan, Shanxi 030006, China) |
| Abstract: | We review applications of large language models (LLMs) in economic and financial research, focusing on text analysis and variable construction, forecasting and decision support, simulation of economic agents, and research assistance and automation. We examine what researchers use LLMs to do, how the reported results compare with existing methods, and what limits their use. Contextual interpretation improves several classification tasks, but variable construction and measurement error affect subsequent analysis. Some forecasts improve on benchmarks, although gains vary and predictive signals need not yield trading profits. Human observations help calibrate agents and improve predictions of survey responses. Automated systems generate economic mechanisms and connect empirical analysis with drafting, while researchers still judge questions and explanations. We then discuss future directions for improving research quality and making better use of researchers' expertise. |
| Keywords: | Large language models, text analysis, economic forecasting, economic agent simulation, research assistance and automation |
| Date: | 2026–09 |
| URL: | https://d.repec.org/n?u=RePEc:kan:wpaper:202618 |