REVIEW 4 major objections 5 minor 18 references
Understanding the Impact of News Articles on the Movement of Market Index: A Case on Nifty 50
T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash
Pith's one-line read Topic-split news sentiment sharpens NIFTY 50 next-day forecasts.
desk verdict A serious data-collection effort on topic-sentiment for NIFTY 50, but the weekend forward-fill and missing significance tests mean the 'impact' claim doesn't stand. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central mechanism is a topic-conditioned sentiment pipeline: each day's news articles are assigned to one of 22 topics by parsing the article URL, headline and full-text sentiments are scored separately (DistilBERT for headlines, VADER for full text), daily per-topic scores are left-joined to NIFTY 50 OHLC data with forward-fill for weekends, and lagged versions of all features are fed to regularized linear regression. The comparative machinery is RMSE on a held-out test period with and without sentiment features, with the largest positive Ridge coefficients used to attribute influence to specific topics.
What would settle it
Re-run the analysis with weekend news sentiments carried to the next trading day (Monday) before aligning with NIFTY data, instead of forward-filling Friday's OHLC values onto weekend rows, and check whether the RMSE advantage of sentiment features survives; if the advantage disappears or flips, the reported improvement is an artifact of the alignment.
Extended reading notes
Core claim
On the paper's own terms, the discovery is that topic-conditioned sentiment scores measurably improve one-day-ahead forecasts of the NIFTY 50 closing value, and that the improvement is largest when headline sentiments come from DistilBERT rather than VADER. In the best configuration (Ridge regression, lag 3, DistilBERT headline scores), the test RMSE drops from 206.01 without sentiment features to 128.74 with them. The fitted coefficients place market lag features and politics_and_nation lag features among the top positive contributors, which the authors take as evidence that both market news and political/national news significantly affect index movement. The paper frames this as closing a gap: prior work used overall sentiment or stock-specific news, while this study disaggregates sentiment by 22 news topics.
Load-bearing premise
The result depends on forward-filling Friday's 'Open', 'High', 'Low', and 'Close' values onto Saturday and Sunday, so that weekend news sentiment is aligned with an unchanged Friday target, which can create predictive signal that would not exist if news were matched to the next trading day's actual movement.
Editorial extensions
If this is right
- Topic-specific news sentiment, not just overall sentiment, directly improves one-day-ahead forecasts of NIFTY 50 closes, with the largest gains from headline-level DistilBERT scores.
- News about markets and about politics/nation carry the strongest positive influence on index movement, as shown by the top Ridge coefficients.
- Regularized linear models such as Ridge with lag-3 sentiment features produce better test RMSE than no-sentiment baselines, while Lasso performs poorly in this setting.
- The findings apply to NIFTY 50 in the study period and should not be generalized to other indices or stocks without fresh analysis.
Reading between the lines
- If the forward-fill alignment is corrected, the reported RMSE gap may shrink, because weekend rows pair unchanged Friday closes with weekend sentiment, giving the model a near-free 'no change' anchor that the no-sentiment baseline does not have.
- The method of inferring significance from the magnitude of Ridge coefficients, without confidence intervals or a significance test, leaves open the possibility that other topics with smaller coefficients are equally predictive; an ablation that removes market/politics topics while retraining would test the attribution.
- The same topic-conditioned pipeline could be extended to other indices and to intraday data, where weekend alignment issues disappear and the impact of political news on market open gaps could be measured directly.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper investigates whether topic-specific sentiment scores from news articles (VADER for full news texts, DistilBERT for headlines) can improve one-day-ahead prediction of the NIFTY 50 index. After scraping roughly 400,000 Economic Times articles from January 2021 to February 2024, the authors group sentiment by 22 topics, create lag-3 and lag-5 feature sets, and compare linear regression, Ridge, Lasso, and Elastic Net with and without sentiment features. They report RMSE values in Tables 1 and 3 and, from the top ridge coefficients in Tables 2 and 4, conclude that market and politics news have significant impact on index movement.
Significance. If established, the topic-specific sentiment decomposition would be a useful extension of news-driven index prediction, suggesting that political and market news should be weighted separately rather than pooled into an aggregate sentiment score. The paper's strengths are the large scraped corpus, the simple URL-based topic extraction, and the use of two sentiment models. However, the empirical support for the headline claim rests on point RMSE differences and unquantified ridge coefficients, and the manuscript contains no code or data availability statement. As detailed below, the weekend forward-filling and absence of significance testing mean the current evidence does not support the conclusion as stated.
major comments (4)
- [Section IV, data preparation paragraph] The left join of the sentiment dataset with the NIFTY dataset, followed by forward-filling of Open, High, Low, and Close, places Saturday and Sunday rows in the training and test sets with Close equal to the preceding Friday's Close. On those rows, the lag-1 Close feature is exactly equal to the target, and the target is not a genuine one-day-ahead market movement. A ridge model can therefore recruit weekend sentiment features as a proxy for the repeated Friday Close, so the with-sentiment RMSE advantage in Tables 1 and 3 may be an artifact of the duplicated target rather than evidence that sentiment moves the index. The authors should repeat the analysis excluding weekend sentiment rows or aligning each news day's sentiment to the next trading day's return, and report whether the RMSE comparisons survive.
- [Abstract and conclusions vs. Tables 1 and 3] The paper uses the word 'significant' throughout (abstract, Section V) but reports only point RMSE values. There are no confidence intervals, permutation tests, or forecast-accuracy tests (e.g., Diebold-Mariano). Similarly, Tables 2 and 4 present the top ridge coefficients without standard errors, bootstrap intervals, or stability analysis; ranking point estimates of regularized coefficients does not establish that market and politics topics have a larger impact. Without uncertainty quantification, the central claim is not statistically supported.
- [Tables 1 and 3] The evidence is internally inconsistent. The 'without sentiment' baselines differ between the two tables for the same model (Ridge: 203.78 vs. 206.01; Linear: 134.85 vs. 137.76), and the VADER-based Elastic Net with sentiment gives RMSE 395.56 versus 174.81 without sentiment, which directly contradicts the text's assertion that regularized regression with sentiment scores 'produced better results' than without. The authors need to reconcile these numbers and specify exactly which baseline configuration is used in each comparison, and they should discuss why sentiment sometimes degrades performance.
- [Section IV, top-feature interpretation] The conclusion that market and politics news have 'significant impact' is derived by reading the signs and magnitudes of coefficients from the same fitted model that produced the RMSE. This is not an independent test of impact; a coefficient in a high-dimensional regularized regression is not a causal effect, and no ablation or external validation (e.g., removing or permuting one topic's sentiment and measuring the change in prediction error) is provided. The authors should supply a falsifiable test of topic-level impact rather than relying on the fitted coefficient ranking.
minor comments (5)
- [Equation (1)] The smoothing parameter alpha is said to have a typical value of 15, but no source is given; please provide a citation to the VADER formula or state the exact implementation used.
- [Section IV, dataset description] The text says 'Total number of data points in each dataset was above 1000' but the period spans more than three years; given the September 2023 removal, the exact count should be stated precisely for reproducibility.
- [Figures 1 and 2] The figures are referenced in the text but no axis labels or caption-level RMSE values are provided, so the reader cannot verify the claimed close match from the figures.
- [General] The manuscript does not state whether the data or code will be made available; a data and code availability statement is needed for reproducibility.
- [Abstract and Section I] There are several grammatical errors, including 'sentiment scores of news items of different other topics' in the abstract and 'the study was done on NIFTY 50 index to understand the impact' in Section I; these do not affect the science but should be corrected.
Circularity Check
Weekend forward-fill makes the target equal to a lagged input feature, so the RMSE gains in Tables 1 and 3 do not establish a significant sentiment impact.
-
self definitional
[Section IV (Results and Analysis), data joining and forward-fill paragraph; supported by Tables 1 and 3]
"the news sentiment dataset was left joined with the NIFTY 50 dataset so that sentiment scores of the Saturdays and Sundays were not removed during joining two datasets. The missing values introduced due to left joining in the variables “Open”, “High”, “Low” and “Close” were filled with forward fill process (i.e., last value carry forward)... The “Close” variable was considered as the target variable."
For Saturday and Sunday rows, forward-fill sets Close equal to the preceding Friday Close. The lagged feature matrix includes lag-1 Close from the immediately preceding row, which is Friday, so on Saturday the target equals the lag-1 feature exactly; on Sunday both lag-1 and lag-2 Close equal Friday's Close. These rows are therefore not one-day-ahead predictions: the target is reconstructible from the model's own inputs by construction. A ridge model with sentiment features can use weekend topic sentiment to compensate for shrinkage of the lagged-Close coefficients, improving the aggregate RMSE in Tables 1 and 3 without any true predictive relationship. The reported sentiment-model advantage is thus partly an artifact of the forward-filled target.
full rationale
The paper is not built on a self-citation chain: the ridge/lasso/Elastic Net comparison is a genuine train/test holdout exercise, and topic-sentiment features are independently extracted from scraped news text. However, the forward-fill of weekend OHLC values creates a constructed equivalence between the target and a lagged feature for Saturday and Sunday rows. Since the aggregate RMSE values in Tables 1 and 3 pool these weekend rows with genuine one-day-ahead weekday predictions, the reported sentiment-model improvements are contaminated and cannot be read as evidence of a significant impact of market or politics news. The conclusion also interprets the largest fitted ridge coefficients as 'influence' without uncertainty quantification, but that is a statistical inference weakness rather than a circular derivation. One concrete by-construction step exists, so the score reflects partial circularity rather than a fully self-referential derivation.
Assumptions & free parameters
free parameters (4)
- VADER alpha =
15
- Topic frequency threshold =
200
- Lag count =
3 and 5
- Regularization hyperparameters =
Grid-searched
assumptions (6)
- domain assumption VADER sentiment scores from news article bodies are meaningful measures of article sentiment.
- domain assumption DistilBERT sentiment on headlines represents the sentiment of the full article.
- domain assumption News topic labels extracted from Economic Times URL paths are correct.
- ad hoc to paper Zero-filling missing topic-day sentiment scores is valid.
- domain assumption Forward-filling OHLC onto weekend dates preserves the temporal relationship between news and index movement.
- domain assumption Linear and ridge regression are adequate models for index movement as a function of lagged features.
Cite this review
Pith. "Pith review of Understanding the Impact of News Articles on the Movement of Market Index: A Case on Nifty 50." pith.science (2026). https://pith.science/paper/AMRO5UXP
@misc{pith2026241206794,
author = {Pith},
title = {Pith review of: Understanding the Impact of News Articles on the Movement of Market Index: A Case on Nifty 50},
year = {2026},
howpublished = {\url{https://pith.science/paper/AMRO5UXP}},
note = {Machine review of arXiv:2412.06794}
}
read the original abstract
In the recent past, there were several works on the prediction of stock price using different methods. Sentiment analysis of news and tweets and relating them to the movement of stock prices have already been explored. But, when we talk about the news, there can be several topics such as politics, markets, sports etc. It was observed that most of the prior analyses dealt with news or comments associated with particular stock prices only or the researchers dealt with overall sentiment scores only. However, it is quite possible that different topics having different levels of impact on the movement of the stock price or an index. The current study focused on bridging this gap by analysing the movement of Nifty 50 index with respect to the sentiments associated with news items related to various different topic such as sports, politics, markets etc. The study established that sentiment scores of news items of different other topics also have a significant impact on the movement of the index.
Reference graph
Works this paper leans on
-
[1]
Stock market volatility, excess returns, and the role of investor sentiment,
W. Y. Lee, C. X. Jiang, and D. C. Indro, “Stock market volatility, excess returns, and the role of investor sentiment,” J Bank Financ, vol. 26, no. 12, pp. 2277–2299, 2002
work page 2002
-
[2]
An effective time series analysis for stock trend prediction using ARIMA model for nifty midcap-50,
B. U. Devi, D. Sundar, and P. Alli, “An effective time series analysis for stock trend prediction using ARIMA model for nifty midcap-50,” International Journal of Data Mining & Knowledge Management Process, vol. 3, no. 1, p. 65, 2013
work page 2013
-
[3]
Sentiment -aware stock market prediction: A deep learning method,
J. Li, H. Bu, and J. Wu, “Sentiment -aware stock market prediction: A deep learning method,” in 2017 international conference on service systems and service management , IEEE, 2017, pp. 1–6
work page 2017
-
[4]
Incorporating stock prices and news sentiments for stock market prediction: A case of Hong Kong,
X. Li, P. Wu, and W. Wang, “Incorporating stock prices and news sentiments for stock market prediction: A case of Hong Kong,” Inf Process Manag, vol. 57, no. 5, p. 102212, 2020
work page 2020
-
[5]
Stock prediction using event -based sentiment an alysis,
M. Makrehchi, S. Shah, and W. Liao, “Stock prediction using event -based sentiment an alysis,” in 2013 IEEE/WIC/ACM International Joint Conferences on Web Intelligence (WI) and Intelligent Agent Technologies (IAT), IEEE, 2013, pp. 337–342
work page 2013
-
[6]
X. Zhang, Y. Li, S. Wang, B. Fang, and P. S. Yu, “Enhancing stock market prediction with extended coupled hidden Markov model over multi -sourced data,” Knowl Inf Syst, vol. 61, pp. 1071–1090, 2019
work page 2019
-
[7]
Accurate stock price forecasting using robust and optimized deep learning models,
J. Sen and S. Mehtab, “Accurate stock price forecasting using robust and optimized deep learning models,” in 2021 International Conference on Intelligen t Technologies (CONIT), IEEE, 2021, pp. 1–9
work page 2021
-
[8]
Stock prediction based on optimized LSTM and GRU models,
Y. Gao, R. Wang, and E. Zhou, “Stock prediction based on optimized LSTM and GRU models,” Sci Program , vol. 2021, pp. 1–8, 2021
work page 2021
Show all 18 references
-
[9]
How to compare market efficiency? The Sharpe ratio based on the ARMA -GARCH forecast,
L. Liu and Q. Chen, “How to compare market efficiency? The Sharpe ratio based on the ARMA -GARCH forecast,” Financial Innovation , vol. 6, no. 1, p. 38, 2020, doi: 10.1186/s40854-020-00200-6
2020 doi
-
[10]
Deep learning using risk - reward function for stock market prediction,
P. Oncharoen and P. Vateekul, “Deep learning using risk - reward function for stock market prediction,” in Proceedings of the 2018 2nd International Conference on Computer Science and Artificial Intelligence , 2018, pp. 556–561
2018
-
[11]
Adaptive stock trading s trategies with deep reinforcement learning methods,
X. Wu, H. Chen, J. Wang, L. Troiano, V. Loia, and H. Fujita, “Adaptive stock trading s trategies with deep reinforcement learning methods,” Inf Sci (N Y), vol. 538, pp. 142–158, 2020
2020
-
[12]
Stock market prediction using machine learning classifiers and social media, news,
W. Khan, M. A. Ghazanfar, M. A. Azam, A. Karami, K. H. Alyoubi, and A. S. Alfakeeh, “Stock market prediction using machine learning classifiers and social media, news,” J Ambient Intell Humaniz Comput, pp. 1–24, 2022
2022
-
[13]
Using social network sentiment analysis and genetic algorithm to improve the stock prediction accuracy of the deep learning- based approach,
J.-Y. Huang, C. -L. Tung, and W. -Z. Lin, “Using social network sentiment analysis and genetic algorithm to improve the stock prediction accuracy of the deep learning- based approach,” International Journal of Computational Intelligence Systems, vol. 16, no. 1, p. 93, 2023
2023
-
[14]
A local and global event sentiment based efficient stock exchange forecasting using deep learning,
H. Maqsood et al. , “A local and global event sentiment based efficient stock exchange forecasting using deep learning,” Int J Inf Manage, vol. 50, pp. 432–451, 2020
2020
-
[15]
Vader: A parsimonious rule-based model for sentiment analysis of social media text,
C. Hutto and E. Gilbert, “Vader: A parsimonious rule-based model for sentiment analysis of social media text,” in Proceedings of the international AAAI conference on web and social media, 2014, pp. 216–225
2014
-
[16]
Attention is all yo u need,
A. Vaswani et al., “Attention is all yo u need,” Adv Neural Inf Process Syst, vol. 30, 2017
2017
-
[17]
Bert: Pre-training of deep bidirectional transformers for language understanding,
J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “Bert: Pre-training of deep bidirectional transformers for language understanding,” arXiv preprint arXiv:1810.04805, 2018
2018 arXiv
-
[18]
DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter,
V. Sanh, L. Debut, J. Chaumond, and T. Wolf, “DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter,” arXiv preprint arXiv:1910.01108, 2019
1910 arXiv
Reviewed August 12, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.