Paper: arXiv 2405.10449

Authors: David Ardia, Keven Bluteau

Abstract

We propose an approach to construct text-based time-series indices in an optimal way–typically, indices that maximize the contemporaneous relation or the predictive performance with respect to a target variable, such as inflation. We illustrate our methodology with a corpus of news articles from the Wall Street Journal by optimizing text-based indices focusing on tracking the VIX index and inflation expectations. Our results highlight the superior performance of our approach compared to existing indices.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 8.5/10
  • Quadrant: Holy Grail — high math complexity, high empirical rigor

Why this score: The paper employs advanced optimization techniques (genetic algorithms with domain-specific operators) and formal mathematical notation for selection matrices, indicating high mathematical complexity. It also demonstrates rigorous empirical validation with a large corpus, out-of-sample testing, and comparative performance against established benchmarks.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Create Optimal Text-Based Indices<br>to Track VIX & Inflation"] --> B["Data Input<br>Wall Street Journal Corpus"]
  B --> C["Methodology<br>Optimization Algorithm<br>Maximizes Contemporaneous/Predictive Power"]
  C --> D["Computational Process<br>Text Processing<br>Feature Extraction<br>Statistical Optimization"]
  D --> E["Comparison<br>vs. Existing Baseline Indices"]
  E --> F{"Key Findings/Outcomes"}
  F --> G["Superior Tracking Performance<br>(VIX & Inflation)"]
  F --> H["Optimal Index Construction<br>Methodology"]

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 an optimal construction of text‑based time‑series indices that maximize contemporaneous or predictive performance for targets like inflation.

Source sentence
We propose an approach to construct text-based time-series indices in an optimal way—typically, indices that maximize the contemporaneous relation or the predictive performance with respect to a target variable, such as inflation.

Assumes 5

  • A simple linear relationship between the target variable and the text‑based factor is assumed for exposition. model
    Source sentence
    To ease the presentation, we focus on a simple linear relationship without additional explanatory variables.
  • The vocabulary is unrestrictive and large; active token selection is optimized jointly with regression parameters. data
    Source sentence
    In optimization problem (6), we assume an unrestrictive and large vocabulary and optimize active tokens selection in addition to the regression parameters.
  • The target series (VIX, inflation expectations) are standardized to avoid scale‑dependent penalty parameters. statistical
    Source sentence
    We standardize yt to avoid scale-dependent penalty parameters.
  • Penalty term λ1 prevents the same token from activating across multiple dimensions in the selection matrix. model
    Source sentence
    I[ωvi′K > 1], which aims to avoid the activation of the same token across multiple dimensions in the selection matrix.
Show 1 more
  • Genetic algorithm with custom crossover and mutation operators can effectively solve the non‑linear minimization problem. execution
    Source sentence
    Below, we present our strategy based on a genetic algorithm to perform the minimization.

Data

  • Wall Street Journal news articles · Wall Street Journal text-news commercial
    Source sentence
    We illustrate our methodology with a corpus of news articles from the Wall Street Journal by optimizing text-based indices focusing on tracking the VIX index and inflation expectations.
  • VIX index · FRED (Federal Reserve Economic Data) macro public
    Source sentence
    The VIX is retrieved from https://fred.stlouisfed.org/series/VIXCLS.
  • MICH inflation expectations index · FRED (Federal Reserve Economic Data) macro public
    Source sentence
    MICH is retrieved from https://fred.stlouisfed.org/series/MICH.

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • The optimal text‑based index reduces RMSE for VIX by 49% in the training window, 39% in validation, and 37% in the test window relative to benchmarks. performance
    Source sentence
    The reduction in RMSE is substantial for the VIX (49% in the
  • For inflation expectations, the optimal index lowers RMSE by 23% in training, 27% in validation, and 25% in the test window versus benchmarks. performance
    Source sentence
    inflation expectations (23% in the training window and 27% in the validation window)
  • Most selected tokens have intuitive relations with the target variable, confirming the data‑driven selection’s relevance. statistical
    Source sentence
    We see that most selected tokens by our optimization strategy have an intuitive relation with the target variable.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 11 related papers). Coverage is partial, so this is neither replication nor refutation.

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.

Cited References (88)

  1. GA : A Package for Genetic Algorithms in R (2013)
  2. On Some Extensions to GA Package: H ybrid Optimisation, Parallelisation and Islands Evolution (2017)
  3. Questioning the news about economic growth: Sparse forecasting using thousands of news-based sentiment values (2019)
  4. Thirty years of academic finance (202x)
  5. Climate Change Concerns and the Performance of Green vs. Brown Stocks (2023)
  6. Measuring Economic Policy Uncertainty (2016)
  7. The structural topic model and applied social science (2013)
  8. Democracy as a Latent Variable (2008) — link
and 80 more
  1. Policy news and stock market volatility (2019)
  2. News implied volatility and disaster concerns (2017)
  3. Econometrics meets sentiment: An overview of methodology and applications (2020)
  4. Realised volatility forecasting: Machine learning via financial word embedding (2021)
  5. News-driven inflation expectations and information rigidities (2021)
  6. Media abnormal tone, earnings announcements, and the stock market (2022)
  7. Text selection (2021)
  8. Questioning the news about economic growth: Sparse forecasting using thousands of news-based sentiment values (2019)
  9. Word power: A new approach for content analysis (2013)
  10. Text as data (2019)
  11. News on inflation and the epidemiology of inflation expectations (2013)
  12. The demand and supply of information about inflation (2022)
  13. Introduction to evolutionary algorithms (2010)
  14. Economic policy uncertainty in the euro area: an unsupervised machine learning approach (2020)
  15. Glove: Global vectors for word representation (2014)
  16. Regression shrinkage and selection via the lasso (1996)
  17. Measuring geopolitical risk (2022)
  18. Measuring climate policy uncertainty (2021)
  19. A century of Economic Policy Uncertainty through the F rench– C anadian lens (2021)
  20. Monetary policy uncertainty (2020)
  21. Trade policy uncertainty (2022)
  22. Measuring economic policy uncertainty (2016)
  23. Structural Topic Models for Open-Ended Survey Responses (2014) — link
  24. Exploring Topic-Metadata Relationships with the STM: A Bayesian Approach (2021) — link
  25. Flexible Arithmetic Asian Options (1995)
  26. An analysis of finance journal accessibility: Author inclusivity and journal quality (2022)
  27. Dissemination, publication, and impact of finance research: When novelty meets conventionality (202x)
  28. Gender representation in economics across topics and time: Evidence from the NBER summer institute (2017)
  29. Comparison and benchmark of name-to-gender inference services (2018)
  30. Are men and women-economists evenly distributed across research fields? Some new empirical evidence (2012)
  31. Gender distribution across topics in the top five economics journals: a machine learning approach (2021)
  32. Beta Regression for Modelling Rates and Proportions (2004) — link
  33. Beta Regression in R (2010) — link
  34. Where should I publish to get promoted? A finance journal ranking based on business school promotions (2020) — link
  35. Just-In-Time M onte C arlo for Path-Dependent A merican Options (2008) — link
  36. American option valuation using first-passage densities (2013)
  37. Correlated topic models (2006)
  38. Text as Data: T he Promise and Pitfalls of Automatic Content Analysis Methods for Political Texts (2013) — link
  39. Latent D irichlet allocation (2003)
  40. Probabilistic topic models (2012)
  41. Probabilistic latent semantic indexing (1999)
  42. A model of text for experimentation in the social sciences (2016) — link
  43. Variational Inference in Nonconjugate Models (2013)
  44. stm : A n R package for structural topic models (2019) — link
  45. The SMART system (1971)
  46. Snowball: A language for stemming algorithms (2001)
  47. Dynamic joint sentiment-topic model (2014)
  48. Introduction to information retrieval (2008)
  49. Web of science (2020)
  50. Key Topics in environmental sociology, 1990–2014: R esults from a computational text analysis (2018) — link
  51. udpipe: Tokenization, Parts of Speech Tagging, Lemmatization and Dependency Parsing with the ‘UDPipe’ ‘NLP’ Toolkit (2021) — link
  52. Tokenizing, pos tagging, lemmatizing and parsing ud 2.0 with udpipe (2017)
  53. Comparing apples to apple: T he effects of stemmers on topic models (2016) — link
  54. Low-dimensional embeddings for interpretable anchor-based topic inference (2014)
  55. Evaluation methods for topic models (2009)
  56. Optimizing semantic coherence in topic models (2011)
  57. Women in the boardroom and their impact on governance and performance (2009)
  58. A Gold Standard Dependency Corpus for English. (2014)
  59. Universal dependencies v2: An evergrowing multilingual treebank collection (2004) — link
  60. The role of prepayment penalties in mortgage loans (2017)
  61. Measuring readability in financial disclosures (2014)
  62. Textual Machine Learning: A n Application to Computational Economics Research (2021) — link
  63. tidytext: Text Mining and Analysis Using Tidy Data Principles in R (2016) — link
  64. An analysis of the intellectual structure of research on the financial economics of precious metals (2019) — link
  65. R: A Language and Environment for Statistical Computing (2020) — link
  66. future.apply: Apply Function to Elements in Parallel using Futures (2020) — link
  67. The igraph software package for complex network research (2006) — link
  68. Transparency and deliberation within the FOMC : A computational linguistics approach (2018) — link
  69. Technical terminology: S ome linguistic properties and an algorithm for identification in text (1995) — link
  70. More efficient topic modelling through a noun only approach (2015)
  71. Automatic Keyword Extraction from Individual Documents (2010) — link
  72. Latent dirichlet allocation for tag recommendation (2009)
  73. Mining meaning from online ratings and reviews: Tourist satisfaction analysis using latent dirichlet allocation (2017)
  74. Using structural topic modeling to detect events and cluster T witter users in the U krainian crisis (2015) — link
  75. Social media approaches to modeling wildfire smoke dispersion: S patiotemporal and social scientific investigations (2017) — link
  76. May it please the T witterverse: T he use of T witter by state high court judges (2019) — link
  77. A correlated topic model of science (2007)
  78. What topic modeling could reveal about the evolution of economics (2018) — link
  79. Citations and the readers’ information-extracting costs of finance articles (2021) — link
  80. Two-stage topic modelling of scientific publications: A case study of University of Nairobi, Kenya (2021)

0 of 88 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.