Paper: arXiv 2411.09937

Authors: Masahiro Suzuki, Hiroki Sakaji

Abstract

We aim to enhance a price sentiment index and to more precisely understand price trends from the perspective of not only consumers but also businesses. We extract comments related to prices from the Economy Watchers Survey conducted by the Cabinet Office of Japan and classify price trends using a large language model (LLM). We classify whether the survey sample reflects the perspective of consumers or businesses, and whether the comments pertain to goods or services by utilizing information on the fields of comments and the industries of respondents included in the Economy Watchers Survey. From these classified price-related comments, we construct price sentiment indices not only for a general purpose but also for more specific objectives by combining perspectives on consumers and prices, as well as goods and services. It becomes possible to achieve a more accurate classification of price directions by employing a LLM for classification. Furthermore, integrating the outputs of multiple LLMs suggests the potential for the better performance of the classification. The use of more accurately classified comments allows for the construction of an index with a higher correlation to existing indices than previous studies. We demonstrate that the correlation of the price index for consumers, which has a larger sample size, is further enhanced by selecting comments for aggregation based on the industry of the survey respondents.

Complexity vs Empirical Score

  • Math Complexity: 2.5/10
  • Empirical Rigor: 7.0/10
  • Quadrant: Street Traders — empirical and applied, lower math complexity

Why this score: The paper employs practical, data-driven NLP techniques (LLMs) and focuses on building indices with high empirical validation (correlation metrics against established CPI/PPI), but lacks advanced mathematical derivations or statistical modeling.

Research Flowchart

  flowchart TD
  A["Research Goal: Enhance price sentiment indices<br>using Economy Watchers Survey comments"] --> B
  
  subgraph B ["Data & Preprocessing"]
      direction LR
      B1["Survey Comments<br>with Industry Metadata"] --> B2["Extract Price-Related Comments"]
  end

  B --> C{"LLM Classification"}
  C --> D["Classify Perspective:<br>Consumer vs. Business"]
  D --> E["Classify Sector:<br>Goods vs. Services"]

  subgraph F ["Sentiment Aggregation"]
      direction LR
      F1["Construct Indices:<br>General, Consumer, Business"] --> F2["Industry-Selection<br>for Consumer Samples"]
  end

  E --> F

  F --> G["Key Findings & Outcomes"]
  G --> H["Higher correlation to<br>existing price indices"]
  G --> I["Potential performance gain<br>via multiple LLM integration"]

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 a framework that uses multiple large language models to extract price‑related comments and build segmented price sentiment indices for consumers and corporations, goods and services.

Source sentence
We propose a framework that integrates multiple LLMs and develops more segmented PSIs from comments classified from the perspectives of consumer and business, goods and services

Assumes 2

  • Comments from the Economy Watchers Survey after 2000 are sufficient for price‑direction analysis, so data from 2001 onward are used. data
    Source sentence
    Since the number of respondents in the survey’s starting year (2000) is significantly lower compared to subsequent years, we use data from 2001 onwards. Out of the total 280,825 entries, 67,060 entries are extracted as being related to prices.
  • Integrating outputs of multiple LLMs can improve classification performance compared with a single model. model
    Source sentence
    A slightly better performance is achieved compared to using only a single model by integrating the outputs of multiple models.

Data

  • Economy Watchers Survey comments · Cabinet Office of Japan macro public
    2001 onward
    Source sentence
    Since the number of respondents in the survey’s starting year (2000) is significantly lower compared to subsequent years, we use data from 2001 onwards. Out of the total 280,825 entries, 67,060 entries are extracted as being related to prices.

Costs

Not applicable The paper reports no trading or execution result.

Finds 3

  • Specific PSI shows a correlation with CPI (Services) about 0.04 higher than the General PSI. performance
    Source sentence
    the correlation between Specific PSI and CPI (Services) is approximately 0.04 higher than that with the General PSI.
  • General PSI Granger‑causes Core CPI (F=42.616, p=0.0000) while the reverse is not significant. statistical
    Source sentence
    General PSI Core Core CPI 42.616*** 0.0000
  • Segmented PSIs have the highest correlation with three consumer‑related price indices compared to previous studies. performance
    Source sentence
    showed the highest correlation with the three consumer-related price indices compared to the previous studies

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 22 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 (41)

  1. Economy Watchers Survey provides Datasets and Tasks for Japanese Financial Domain (2024) — link
  2. Extracting firms’ short-term inflation expectations from the Economy Watchers Survey using text analysis (2021)
  3. S udachi: a J apanese Tokenizer for Business (2018) — link
  4. Constructing and analyzing domain-specific language model for financial text mining (2023) — link
  5. FinDeBERTaV2: Word-Segmentation-Free Pre-trained Language Model for Finance (2024) — link
  6. The causal relationship between producer price index and consumer price index: Empirical evidence from selected European countries (2011)
  7. Dynamic causality between PPI and CPI in China: A rolling window bootstrap approach (2023)
  8. Exchange rate pass-through to Japanese prices: Import prices, producer prices, and the core CPI (2022) — link
and 33 more
  1. Forecasting CPI inflation components with Hierarchical Recurrent Neural Networks (2023) — link
  2. Forecasting UK inflation bottom up (2024) — link
  3. Internet search behavior as an economic forecasting tool: The case of inflation expectations (2011)
  4. Estimation of firms’ inflation expectations using the survey DI (2023)
  5. Nowcasting prices using Google trends: an application to Central America (2015)
  6. News-based business sentiment and its properties as an economic index (2022) — link
  7. Forecasting Japanese inflation with a news-based leading indicator of economic activities (2021) — link
  8. Are ChatGPT and GPT-4 General-Purpose Solvers for Financial Text Analytics? A Study on Several Typical Tasks (2023) — link
  9. Measuring news sentiment (2022) — link
  10. Predicting Socio-Economic Indicators using News Events (2016) — link
  11. Is C hat GPT a Financial Expert? Evaluating Language Models on Financial Natural Language Processing (2023) — link
  12. Stock Selection Attempt using Sentiment of Japan Company Handbook (2024) — link
  13. Sentiment Summarization of Financial Reports by LSTM RNN model with the Japan Economic Watcher Survey Data (2016) — link
  14. FinBen: A Holistic Financial Benchmark for Large Language Models (2024) — link
  15. Construction of a J apanese Financial Benchmark for Large Language Models (2024) — link
  16. Economic analysis using machine learning: Text mining of the Economy Watchers Survey (2018)
  17. Extracting firms’ short-term inflation expectations from survey comments using text analysis (2022)
  18. Can ChatGPT improve investment decisions? From a portfolio management perspective (2024) — link
  19. Can ChatGPT assist in picking stocks? (2024) — link
  20. Chatgpt-based investment portfolio selection — → 2308.06260
  21. Learning to Generate Explainable Stock Predictions using Self-Reflective Large Language Models — → 2402.03659
  22. LLMF actor: Extracting Profitable Factors through Prompts for Explainable Stock Movement Prediction (2024) — link
  23. The consumer price index prediction using machine learning approaches: Evidence from the United States (2023)
  24. Comparison four kernels of svr to predict consumer price index (2021)
  25. Can we measure inflation expectations using Twitter? (2022) — link
  26. LLM-jp: A Cross-organizational Project for the Research and Development of Fully Open Japanese LLMs (2024) — link
  27. From Base to Conversational: Japanese Instruction Dataset and Tuning Large Language Models (2023) — link
  28. Continual Pre-Training for Cross-Lingual LLM Adaptation: Enhancing Japanese Language Capabilities (2024)
  29. JaFIn: Japanese Financial Instruction Dataset (2024) — link
  30. Development and analysis of medical instruction-tuning for Japanese large language models (2024) — link
  31. Large Language Models are legal but they are not: Making the case for a powerful LegalLLM (2023) — link
  32. Language models are few-shot learners (2020)
  33. Large language models are zero-shot reasoners (2022)

2 of 41 cited works are in the corpus; resolved via arXiv source bundles, Semantic Scholar and OpenAlex. Corpus-wide: 2596 linked papers.