Pith. sign in

REVIEW 5 major objections 5 minor 77 references

Stock Market Telepathy: Graph Neural Networks Predicting the Secret Conversations between MINT and G7 Countries

T0 review · 5 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A graph neural network called MTGNN forecasts MINT and G7 stock indices more accurately than four standard baselines, and the network's learned links identify the most influential markets in each bloc.

desk verdict A new application of MTGNN to MINT/G7 stock indices with an honest non-causality caveat, but the forecasting protocol is too underspecified to support the headline outperformance claim. read the letter →

arxiv 2506.01945 v1 pith:KO6GNUF7 submitted 2025-06-02 econ.EM cs.LGstat.AP

Nurbanu Bursa This is my paper
classification econ.EMcs.LGstat.AP
keywords graphneuralnetworksMTGNNstockindexforecastingMINTcountriesG7multivariatetimeseriesfinancialinterconnectednessdeeplearning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper claims that a graph neural network designed for multivariate time series, MTGNN, can forecast the daily closing prices of the main stock indices of the MINT and G7 countries more accurately than four standard alternatives (AR, VAR-MLP, RNN-GRU, TCN). Applying the model to 2012–2024 data, the author reports that MTGNN achieves the lowest error on all eleven indices, with particularly large gains on the more volatile emerging-market indices. The learned graph structure is then read as a map of influence: column sums of the adjacency matrix indicate that the US and Canada are the most influential G7 indices and Indonesia and Türkiye the most influential MINT indices in the forecasting process. The paper's significance is that if correct, it offers a data-driven way to model financial interconnectedness between economic blocs without pre-specifying the links, and a forecasting tool for markets that are hard to predict.

What carries the argument

The machinery is MTGNN, a multivariate time series forecasting network that learns both the graph and the forecast end-to-end. Each stock index is a node; a graph learning layer constructs an adjacency matrix from node embeddings, then interleaved graph convolution and temporal convolution modules propagate information across the learned edges and across time, with residual and skip connections. The learned adjacency matrix is the object that carries the influence analysis: its column sums, counted over 1-hop and 2-hop neighborhoods, rank which countries' indices most affect the others in the forecasting process.

What would settle it

Re-run the MTGNN and all baseline models with normalization statistics (per-column mean and standard deviation) estimated only on the 60% training split, then recompute RSE, RMSE, MAE, and MAPE on the final 20% test split; if MTGNN's reported margins over the best baseline shrink materially or reverse, the claimed out-of-sample outperformance is an artifact of data leakage.

Watch

Extended reading notes

Core claim

The central claim is that a spatio-temporal graph neural network, MTGNN, outperforms the classical and deep-learning baselines in forecasting stock index levels for MINT and G7 countries, and that the graph it learns encodes meaningful cross-country influence. Using daily closing prices of eleven indices (FTSE MIB, BIST 100, CAC 40, FTSE 100, DAX, S&P 500, S&P/TSX, IDX Composite, IPC Mexico, Nikkei 225, NSE 30) from 30 January 2012 to 14 August 2024, the paper trains on 60% of the data, validates on 20%, and tests on the last 20%. On the test period, MTGNN reports the smallest RSE, RMSE, MAE, and MAPE for every index; for example, RSE values lie between 0.013 (Nigeria) and 0.081 (Türkiye), whereas baseline RSEs go as low as 0.045 and as high as 66.671. The influence ranking comes from the learned adjacency matrix: counting out-degree connections in 1-hop and 2-hop neighborhoods, the US and Canada are the most influential G7 countries, and Indonesia and Türkiye the most influential MINT countries in the prediction process. The paper is explicit that this graph is an optimization artifact, not a causal model.

Load-bearing premise

The entire outperformance claim depends on the test set being genuinely out-of-sample: if the per-column mean and standard deviation used for normalization in Remark 1 are calculated from the full series rather than from the training portion only, the test data enters the model through the scaler and the reported errors are optimistically biased.

Editorial extensions

If this is right

  • If MTGNN's accuracy is genuine, it provides a practical forecasting tool for volatile emerging-market indices, where the best baseline errors are several times larger.
  • The learned adjacency matrix can serve as a data-driven alternative to correlation and dynamic time warping for measuring financial interconnectedness across economic blocs.
  • The influence ordering suggests that monitoring US and Canadian market movements is informative for forecasting other G7 and MINT indices, and that Indonesian and Turkish indices are the most informative within MINT.
  • The same end-to-end graph-learning pipeline can be applied to other country groupings or asset classes whenever the dependency structure is unknown.
  • Because the graph is learned rather than fixed, the model can be retrained on rolling windows to track changes in cross-country financial influence over time.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper does not test what the learned graph contributes beyond the architecture; ablating the graph (replacing learned edges with random or fully connected graphs) would reveal how much of MTGNN's gain comes from the graph structure itself.
  • The influence ranking is read off raw out-degree counts in the learned adjacency matrix; a weighted or PageRank-style centrality measure, or a Granger-causality check on forecast errors, could produce a different and potentially more reliable ordering.
  • The test window (January 2022 to August 2024) covers a period of post-COVID inflation and rapid interest-rate changes, so the reported error margins may not hold in calmer regimes; a rolling-window evaluation would test stability.
  • The 'secret conversations' metaphor implies information transmission, but the model only captures statistical dependence for level forecasting; applying the same model to volatility or risk measures would show whether the learned edges represent risk spillovers rather than level co-movement.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper applies the MTGNN graph neural network to daily closing prices of eleven stock market indices (G7 and MINT countries) from January 30, 2012 to August 14, 2024. It reports that MTGNN outperforms AR, VAR-MLP, RNN-GRU, and TCN on RSE, RMSE, MAE, and MAPE (Table 4), and interprets the learned adjacency matrix as revealing that the US and Canada (among G7) and Indonesia and Türkiye (among MINT) are the most influential countries in the forecasting process. The paper frames this as the first application of a spatio-temporal GNN to economic blocs and emphasizes the interpretability of the learned graph structure.

Significance. If the results are valid, the paper offers a novel empirical application of a recent graph-based deep learning architecture to cross-country stock index forecasting, with the additional interpretability angle that the learned graph structure can identify influential countries. The author is to be credited for using a known and reproducible architecture, for providing the learned adjacency matrix, and for including an explicit disclaimer that the learned graph should not be interpreted as causal. However, the empirical claims currently rest on a single table with no uncertainty quantification and on an under-specified evaluation protocol, so the paper's contribution is not yet established at the level required for a finance/econometrics journal.

major comments (5)
  1. [Section 4.1.1 and Section 5] The forecasting task is not fully specified. Equation (1) in Section 4.1.1 defines the prediction target as Y = {z_{tP+1}, ..., z_{tP+Q}} but Q is never assigned a value, and Section 5 does not state whether the test-period evaluation uses one-step-ahead forecasts with observed lags, recursive multi-step forecasts, or direct multi-step forecasts. Without this definition, the numbers in Table 4 cannot be interpreted as coming from a single well-defined forecasting problem, and the claimed outperformance is not reproducible.
  2. [Remark 1 and Section 5] The normalization protocol is ambiguous about whether the column means and standard deviations used for normalization are computed on the training segment only or on the full dataset. If they are computed on the full series, the test set is not independent and the reported errors (e.g., RSE 0.013–0.081) are optimistically biased. The statement in Remark 1 that transformations are 'inverted for the construction of the plots' makes the distinction between training and test statistics even more consequential; the paper must state explicitly that all normalization statistics come from the training portion only, and ideally provide the exact transformation pipeline.
  3. [Table 4] Table 4 reports point estimates without any measure of uncertainty (standard errors, confidence intervals, or multiple-seed variability), and several entries indicate likely typographical errors. For example, the RNN-GRU row for FTSE MIB reports MAE = 1.005 alongside RMSE = 0.148, which is impossible because MAE cannot exceed RMSE. The TCN row for IDX COMPOSITE reports MAPE as '2.61%' while other cells use one decimal format, and the RNN-GRU row for FTSE MIB appears to have misplaced values. The authors should correct these anomalies, add uncertainty quantification or at least multiple-seed results, and explain why MTGNN's advantage is so large (e.g., RSE 0.026 vs. next-best 1.005) that it raises concerns about a mismatch in evaluation protocol or baseline tuning.
  4. [Section 5, adjacency matrix and conclusions] The 'most influential' claim is based on the learned adjacency matrix A, which is fitted to minimize training loss. The paper correctly disclaims causality in the paragraph following Equation (5), but the abstract and conclusions restate the out-degree analysis of A and A+A^2 as evidence that the US, Canada, Indonesia, and Türkiye are 'the most influential' countries. This is a post hoc reading of fitted parameters; without stability analysis (e.g., bootstrap over training windows or random seeds) or ablations (e.g., removing edges and measuring forecast degradation), the influence claim is not supported as an empirical finding.
  5. [Section 4.2 and Table 4] The baseline implementations are described only verbally. The paper does not report the lag order for AR, the architecture and hyperparameters for VAR-MLP, RNN-GRU, and TCN, or whether the baselines received the same input features, normalization, and tuning budget as MTGNN. Without these details, the comparison in Table 4 may be unfair if the baselines were not appropriately configured. Please provide complete implementation details or cite a specific benchmark configuration for each baseline.
minor comments (5)
  1. [Figure 1 and Figure 6] The label 'NIKKEI 255' appears in Figure 1 and in the panel title of Figure 6; the correct index name is NIKKEI 225.
  2. [Section 4.1.1] The set notation 'X ⊇ X = {z_t1, ...}' is confusing and likely contains a typo; it should be either 'X ⊂' or simply 'X ='.
  3. [Section 5, data split] The paper states the data run from January 30, 2012 to August 14, 2024 and that the last 20% starts in January 2022, but given the full sample length this implies a test set of about 916 trading days, which is plausible. Please verify that the 60/20/20 split boundaries correspond exactly to the reported dates.
  4. [Section 2, literature review] The literature review paragraph is very long and lists many references without much synthesis; grouping by approach (statistical, machine learning, graph neural networks) would improve readability.
  5. [Throughout] The paper uses the word 'influential' in the abstract and conclusions, but the author's own caveat in Section 5 says the learned graph is not causal; please align the language consistently, for example by using 'strongly connected in the learned graph' or 'important in the fitted model' instead of 'influential'.

Circularity Check

1 steps flagged · score 4.0 of 10

The forecast benchmark is not circular, but the headline 'influential countries' finding reduces by construction to column sums of the Adam-fitted adjacency matrix, so a secondary headline claim is a fitted-parameter summary.

  1. fitted input called prediction [Section 5, text following Eq. (5); Abstract]
    "Performing sums over the columns of the adjacency matrix (5) reveals that if only 1-hop neighborhoods are considered, the US, Germany, and Canada are the most influential stock indices among G7 countries with 7, 5, and 5 out-degree connections, respectively, while among MINT countries, Indonesia and Türkiye are the most influential stock indices with 7 and 6 out-degree connections, respectively, in the forecasting process."

    The headline 'most influential' result is computed as column sums of the adjacency matrix A in Eq. (5), and A is the learnable parameter fitted end-to-end by Adam on the training loss; the paper states the graph is 'constructed by performing and end-to-end stochastic gradient descent ... on the losses accrued between the predicted and the actual values.' Every fitted A necessarily has some out-degree maximum, so the ranking is a by-construction summary of a fitted parameter, not an empirical discovery or a prediction validated on held-out data.

full rationale

The central derivation is not circular: MTGNN is borrowed from an external, non-self-cited source (Wu et al., reference [68]), and the Table 4 forecast errors are computed on the last 20% of the series (test split starting January 2022), so no fitted parameter is renamed as a forecast of held-out data. There is no self-citation chain and no uniqueness theorem imported from the author's prior work, since the paper has a single author and cites no own prior results. The one circular-adjacent step is the secondary 'influential countries' finding, which is obtained by column sums of the learned adjacency matrix and is therefore a by-construction property of a fitted parameter presented in the abstract as a revelation; this is flagged above and contributes the partial circularity. The paper's own limitation statement ('by no means do we claim that the constructed graph in Figure 5, is a causal one') is weighed in mitigation. Remaining concerns are validity issues rather than circularity and do not meet the quoted-reduction bar: the forecast horizon Q in Section 4.1.1 is never stated, the test-time application scheme (one-step-ahead with observed lags versus recursive multi-step) is undefined, Remark 1 does not say whether normalization statistics are restricted to the training segment, and baseline hyperparameters are not documented. These make the benchmark hard to interpret but are not circular by construction. Score 4 reflects partial circularity confined to the influence claim while the central forecast-accuracy claim retains independent out-of-sample content.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the learned adjacency matrix (a fitted object), the predictability assumption, the stationarity of the graph over time, and the unspecified fairness of the baseline comparison. No new entities are introduced.

free parameters (1)
  • Learned adjacency matrix A = 11x11 matrix with entries like 0.98, 0.96, 0.99, 0.9, 0.81, and 1.0 (Eq. 5)
    The out-degree counts and the conclusion about which countries are influential are derived directly from these learned weights, which are optimized on the training data. No threshold or external validation is reported.
assumptions (3)
  • domain assumption Future stock index values are predictable from past values of the same and other indices in the set.
    The entire forecasting exercise depends on this predictability; the paper does not test for unit roots or compare against a no-predictability benchmark.
  • domain assumption The graph structure learned from the training period remains valid in the test period.
    The interpretation of the learned adjacency matrix as 'influence' assumes the training-period dependencies persist out-of-sample and over time.
  • domain assumption Baseline models were implemented and tuned with comparable effort.
    The claim that MTGNN outperforms traditional methods depends on the baselines not being disadvantaged by poor implementation or hyperparameter choices; no details are given.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Stock Market Telepathy: Graph Neural Networks Predicting the Secret Conversations between MINT and G7 Countries." pith.science (2026). https://pith.science/paper/KO6GNUF7

@misc{pith2026250601945,
  author       = {Pith},
  title        = {Pith review of: Stock Market Telepathy: Graph Neural Networks Predicting the Secret Conversations between MINT and G7 Countries},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KO6GNUF7}},
  note         = {Machine review of arXiv:2506.01945}
}
read the original abstract

Emerging economies, particularly the MINT countries (Mexico, Indonesia, Nigeria, and T\"urkiye), are gaining influence in global stock markets, although they remain susceptible to the economic conditions of developed countries like the G7 (Canada, France, Germany, Italy, Japan, the United Kingdom, and the United States). This interconnectedness and sensitivity of financial markets make understanding these relationships crucial for investors and policymakers to predict stock price movements accurately. To this end, we examined the main stock market indices of G7 and MINT countries from 2012 to 2024, using a recent graph neural network (GNN) algorithm called multivariate time series forecasting with graph neural network (MTGNN). This method allows for considering complex spatio-temporal connections in multivariate time series. In the implementations, MTGNN revealed that the US and Canada are the most influential G7 countries regarding stock indices in the forecasting process, and Indonesia and T\"urkiye are the most influential MINT countries. Additionally, our results showed that MTGNN outperformed traditional methods in forecasting the prices of stock market indices for MINT and G7 countries. Consequently, the study offers valuable insights into economic blocks' markets and presents a compelling empirical approach to analyzing global stock market dynamics using MTGNN.

Figures

Figures reproduced from arXiv: 2506.01945 by the authors.

Figure 1
Figure 1. Daily closing price of the stock indices. [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 3
Figure 3. Distance matrix produced by dynamic time warping [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. The architecture of the spatio-temporal graph neu [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: The connectivity of the countries’ stock indices d [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: The predictions obtained MTGNN predictor. [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

77 extracted references · 71 canonical work pages

  1. [1]

    Financial vul nerability and risks to growth in emerging markets (no

    Acharya, V ., Bhadury, S., Surti, J., 2020. Financial vul nerability and risks to growth in emerging markets (no. w27411)

    work page 2020

  2. [2]

    A new stock price forecasting method using active deep learning approach

    Alkhatib, K., Khazaleh, H., Alkhazaleh, H.A., Alsoud, A .R., Abualigah, L., 2022. A new stock price forecasting method using active deep learning approach. Journal of Open Innovation: Technology, Market, and Com- plexity 8, 96

    work page 2022

  3. [3]

    An empirical eva luation of generic convolutional and recurrent networks for sequence modelin g

    Bai, S., Kolter, J.Z., Koltun, V ., 2018. An empirical eva luation of generic convolutional and recurrent networks for sequence modelin g. arXiv preprint arXiv:1803.01271

    arXiv 2018

  4. [4]

    Autoregress ive convolutional neural networks for asynchronous time series, in: Internat ional Confer- ence on Machine Learning, PMLR

    Binkowski, M., Marti, G., Donnat, P ., 2018. Autoregress ive convolutional neural networks for asynchronous time series, in: Internat ional Confer- ence on Machine Learning, PMLR. pp. 580–589

    work page 2018

  5. [5]

    The economist who invented the brics j ust invented a whole new group of countries: The mints

    Boesler, M., 2013. The economist who invented the brics j ust invented a whole new group of countries: The mints. URL: https://www.businessinsider.com/jim-oneill-presents-the-mint-economies-2013-11

    work page 2013

  6. [6]

    Cakra, Y .E., Trisedya, B.D., 2015. Stock price predicti on using linear regression based on sentiment analysis, in: 2015 internati onal confer- ence on advanced computer science and information systems ( ICACSIS), IEEE. pp. 147–154

    work page 2015

  7. [7]

    Mul- tiobjective evolution of fuzzy rough neural network via dis tributed par- allelism for stock prediction

    Cao, B., Zhao, J., Lv, Z., Gu, Y ., Y ang, P ., Halgamuge, S.K ., 2020. Mul- tiobjective evolution of fuzzy rough neural network via dis tributed par- allelism for stock prediction. IEEE Transactions on Fuzzy S ystems 28, 939–952

    work page 2020

  8. [8]

    Chen, L., Chen, D., Shang, Z., Wu, B., Zheng, C., Wen, B., Z hang, W.,

    work page

Show all 77 references
  1. [9]

    A hybrid fuzzy time series m odel based on granular computing for stock price forecasting

    Chen, M.Y ., Chen, B.T., 2015. A hybrid fuzzy time series m odel based on granular computing for stock price forecasting. Informa tion Sciences 294, 227–241

    2015

  2. [10]

    Multi- range attentive bicomponent graph convolutional network f or traffic fore- casting, in: Proceedings of the AAAI conference on artificia l intelligence, pp

    Chen, W., Chen, L., Xie, Y ., Cao, W., Gao, Y ., Feng, X., 20 20. Multi- range attentive bicomponent graph convolutional network f or traffic fore- casting, in: Proceedings of the AAAI conference on artificia l intelligence, pp. 3529–3536

  3. [11]

    A novel graph convo- lutional feature based convolutional neural network for st ock trend pre- diction

    Chen, W., Jiang, M., Zhang, W.G., Chen, Z., 2021. A novel graph convo- lutional feature based convolutional neural network for st ock trend pre- diction. Information Sciences 556, 67–94

    2021

  4. [12]

    Multi-channel fusion graph neu ral network for multivariate time series forecasting

    Chen, Y ., Xie, Z., 2022. Multi-channel fusion graph neu ral network for multivariate time series forecasting. Journal of Computat ional Science 64, 101862

    2022

  5. [13]

    A hybrid model based on rough sets theory and genetic algorithms for stock price forecasting

    Cheng, C.H., Chen, T.L., Wei, L.Y ., 2010. A hybrid model based on rough sets theory and genetic algorithms for stock price forecasting. Information Sciences 180, 1610–1629

    2010

  6. [14]

    Learning phrase representations using r nn encoder-decoder for statistical machine translation

    Cho, K., 2014. Learning phrase representations using r nn encoder-decoder for statistical machine translation. arX iv preprint arXiv:1406.1078

    2014 arXiv

  7. [15]

    Deep learning netw orks for stock market analysis and prediction: Methodology, data represe ntations, and case studies

    Chong, E., Han, C., Park, F.C., 2017. Deep learning netw orks for stock market analysis and prediction: Methodology, data represe ntations, and case studies. Expert Systems with Applications 83, 187–205

    2017

  8. [16]

    Empirical evaluation of gated recurre nt neural networks on sequence modeling

    Chung, J., 2014. Empirical evaluation of gated recurre nt neural networks on sequence modeling. arXiv preprint arXiv:1412.3555

    2014 arXiv

  9. [17]

    Metro: a generic graph neural network framework for multiva riate time series forecasting

    Cui, Y ., Zheng, K., Cui, D., Xie, J., Deng, L., Huang, F., Zhou, X., 2021. Metro: a generic graph neural network framework for multiva riate time series forecasting. Proceedings of the VLDB Endowment 15, 2 24–236

    2021

  10. [18]

    Knowledge-driven stock trend prediction and explanation v ia temporal convolutional network, in: Companion proceedings of the 20 19 world wide web conference, pp

    Deng, S., Zhang, N., Zhang, W., Chen, J., Pan, J.Z., Chen , H., 2019. Knowledge-driven stock trend prediction and explanation v ia temporal convolutional network, in: Companion proceedings of the 20 19 world wide web conference, pp. 678–685

    2019

  11. [19]

    The mint countries as emerging econ omic power bloc: Prospects and challenges

    Durotoye, A., 2014. The mint countries as emerging econ omic power bloc: Prospects and challenges. Developing Country Studie s 4, 1–9

    2014

  12. [20]

    G7 countries’ combined gross domestic prod- uct (gdp) as share of the world’s gdp from 2000 to 2023

    Dyvik, E., 2024. G7 countries’ combined gross domestic prod- uct (gdp) as share of the world’s gdp from 2000 to 2023. URL: https://www.statista.com/statistics/1372791/g7-combined-gdp-share-world/

    2024

  13. [21]

    Economics, W., 2024. G7. URL: https://www.worldeconomics.com/Regions/G7/

    2024

  14. [22]

    E fficient capital markets

    Fama, E.F., 1970. E fficient capital markets. Journal of finance 25, 383– 417

    1970

  15. [23]

    Capital flows, push versus pull fa ctors and the global financial crisis

    Fratzscher, M., 2012. Capital flows, push versus pull fa ctors and the global financial crisis. Journal of International Economic s 88, 341–356

    2012

  16. [24]

    The application of st ock index price prediction with neural network

    Gao, P ., Zhang, R., Y ang, X., 2020. The application of st ock index price prediction with neural network. Mathematical and Computat ional Appli- cations 25, 53

    2020

  17. [25]

    Comprehensive Guide to Dynamic Tim e Warping in Python

    Gulzar, H.M., 2018. Comprehensive Guide to Dynamic Tim e Warping in Python. Lambert Academic Publishing

    2018

  18. [26]

    Graph representation learning

    Hamilton, W.L., 2020. Graph representation learning. Morgan & Clay- pool Publishers

    2020

  19. [27]

    Stock prediction us ing deep learning with long-short-term-memory networks

    Hargreaves, C.A., Leran, C., 2020. Stock prediction us ing deep learning with long-short-term-memory networks. International Jou rnal of Elec- tronic Engineering and Computer Science 5, 22–32

    2020

  20. [28]

    Multivariate time ser ies deep spa- tiotemporal forecasting with graph neural network

    He, Z., Zhao, C., Huang, Y ., 2022. Multivariate time ser ies deep spa- tiotemporal forecasting with graph neural network. Applie d Sciences 12, 5731

    2022

  21. [29]

    Nse stock market prediction using deep-learning models

    Hiransha, M., Gopalakrishnan, E.A., Menon, V .K., Soma n, K., 2018. Nse stock market prediction using deep-learning models. Proce dia computer science 132, 1351–1362

    2018

  22. [30]

    Long short-term memory

    Hochreiter, S., 1997. Long short-term memory. Neural C omputation MIT-Press

    1997

  23. [31]

    St-trader: A spatial- temporal deep neural network for modeling stock market move ment

    Hou, X., Wang, K., Zhong, C., Wei, Z., 2021. St-trader: A spatial- temporal deep neural network for modeling stock market move ment. IEEE/CAA Journal of Automatica Sinica 8, 1015–1024

    2021

  24. [32]

    Investing

    Investing, 2024. Investing. URL: https://www.investing.com/

    2024

  25. [33]

    Forecasting realised volatility using arfima and har models

    Izzeldin, M., Hassan, M.K., Pappas, V ., Tsionas, M., 20 19. Forecasting realised volatility using arfima and har models. Quantitati ve Finance 19, 1627–1638

  26. [34]

    Arima modeling with int ervention to fore- cast and analyze chinese stock prices

    Jarrett, J.E., Kyper, E., 2011. Arima modeling with int ervention to fore- cast and analyze chinese stock prices. International Journ al of Engineer- ing Business Management 3, 17

    2011

  27. [35]

    Mul- tivariate time series forecasting with dynamic graph neura l odes

    Jin, M., Zheng, Y ., Li, Y .F., Chen, S., Y ang, B., Pan, S., 2022. Mul- tivariate time series forecasting with dynamic graph neura l odes. IEEE Transactions on Knowledge and Data Engineering 35, 9168–91 80

    2022

  28. [36]

    Stock price forecast- ing using support vector machines and improved particle swarm optimiza- tion

    Karazmodeh, M., Nasiri, S., Hashemi, S.M., 2013. Stock price forecast- ing using support vector machines and improved particle swarm optimiza- tion. Journal of Automation and control engineering 1, 173– 176

    2013

  29. [37]

    Di ffusion convolutional re- current neural network: Data-driven tra ffic forecasting

    Li, Y ., Y u, R., Shahabi, C., Liu, Y ., 2017. Di ffusion convolutional re- current neural network: Data-driven tra ffic forecasting. arXiv preprint arXiv:1707.01926

    2017 arXiv

  30. [38]

    Multivariate time-series forecasting with tempo ral polynomial graph neural networks

    Liu, Y ., Liu, Q., Zhang, J.W., Feng, H., Wang, Z., Zhou, Z ., Chen, W., 2022. Multivariate time-series forecasting with tempo ral polynomial graph neural networks. Advances in neural information proc essing sys- tems 35, 19414–19426

    2022

  31. [39]

    An integ rated frame- work of deep learning and knowledge graph for prediction of s tock price trend: An application in chinese stock exchange market

    Long, J., Chen, Z., He, W., Wu, T., Ren, J., 2020. An integ rated frame- work of deep learning and knowledge graph for prediction of s tock price trend: An application in chinese stock exchange market. App lied Soft Computing 91, 106205

    2020

  32. [40]

    New Introduction to Multiple Time Series Analysis

    L ¨ utkepohl, H., 2005. New Introduction to Multiple Time Series Analysis. Springer-V erlag

    2005

  33. [41]

    Investor sentiment and th e prediction of stock returns: a quantile regression approach

    Ma, C., Xiao, S., Ma, Z., 2018. Investor sentiment and th e prediction of stock returns: a quantile regression approach. Applied E conomics 50, 5401–5415

    2018

  34. [42]

    Mensi, W., Hammoudeh, S., Reboredo, J.C., Nguyen, D.K. , 2014. Do global factors impact brics stock markets? a quantile regression approach. Emerging Markets Review 19, 1–17

    2014

  35. [43]

    Dynamic time warping in information r etrieval for music and motion

    Muller, M., 2007. Dynamic time warping in information r etrieval for music and motion. Dynamic time warping Information retrieval for music and motion , 69–84

    2007

  36. [44]

    Research guides: Mint: Sources of economic information

    Nagashybayeva, G., 2020. Research guides: Mint: Sources of economic information. URL: https://guides.loc.gov/mexico-indonesia-nigeria-tur key

    2020

  37. [45]

    Multi-step-ahead s tock price predic- tion using recurrent fuzzy neural network and variational m ode decompo- sition

    Nasiri, H., Ebadzadeh, M.M., 2023. Multi-step-ahead s tock price predic- tion using recurrent fuzzy neural network and variational m ode decompo- sition. Applied Soft Computing 148, 110867

    2023

  38. [46]

    World investment report

    Nations, U., 2022. World investment report. URL: https://unctad.org/publication/world-investment-report-2022

    2022

  39. [47]

    The fresh challenge of the mint countrie s

    Newland, 2014. The fresh challenge of the mint countrie s. URL: https://newlandchase.com/the-fresh-challenges-of-th e-mint-countri

    2014

  40. [48]

    Stoc k price prediction using deep learning algorithm and its comparison with machi ne learning algorithms

    Nikou, M., Mansourfar, G., Bagherzadeh, J., 2019. Stoc k price prediction using deep learning algorithm and its comparison with machi ne learning algorithms. Intelligent Systems in Accounting, Finance an d Management 26, 164–174

    2019

  41. [49]

    Stock price predic tion based on 11 error correction model and granger causality test

    Ning, Y ., Wah, L.C., Erdan, L., 2019. Stock price predic tion based on 11 error correction model and granger causality test. Cluster Computing 22, 4849–4858

    2019

  42. [50]

    Mint: meaning, history and econom ic overview

    Okoroafor, R., 2024. Mint: meaning, history and econom ic overview. URL: https://www.makemoney.ng/mint

    2024

  43. [51]

    Wa venet: A generative model for raw audio

    Oord, A.v.d., Dieleman, S., Zen, H., Simonyan, K., Viny als, O., Graves, A., Kalchbrenner, N., Senior, A., Kavukcuoglu, K., 2016. Wa venet: A generative model for raw audio. arXiv preprint arXiv:1609. 03499

    2016

  44. [52]

    Stock price forecast of ma cro-economic factor using recurrent neural network

    Pahlawan, M.R., Riksakomara, E., Tyasnurita, R., Mukl ason, A., Ma- hananto, F., Vinarti, R.A., 2021. Stock price forecast of ma cro-economic factor using recurrent neural network. IAES International Journal of Ar- tificial Intelligence 10, 74

    2021

  45. [53]

    Understanding deep learning

    Prince, S.J., 2023. Understanding deep learning. MIT p ress

    2023

  46. [54]

    Rahmani, S., Baghbani, A., Bouguila, N., Patterson, Z. , 2023. Graph neural networks for intelligent transportation systems: A survey. IEEE Transactions on Intelligent Transportation Systems 24, 88 46–8885

    2023

  47. [55]

    Dilemma not trilemma: the global financia l cycle and monetary policy independence

    Rey, H., 2015. Dilemma not trilemma: the global financia l cycle and monetary policy independence. Technical Report. National Bureau of Economic Research

    2015

  48. [56]

    Stock price predicti on using regression analysis

    Sahoo, P ., Charlapally, K., 2015. Stock price predicti on using regression analysis. International Journal of Scientific & Engineerin g Research 6, 1655–1659

    2015

  49. [57]

    Sawhney, R., Agarwal, S., Wadhwa, A., Shah, R., 2020a. D eep attentive learning for stock movement prediction from social media te xt and com- pany correlations, in: Proceedings of the 2020 Conference o n Empirical Methods in Natural Language Processing (EMNLP), pp. 8415–8 426

    2020

  50. [58]

    Ex ploring the scale-free nature of stock markets: Hyperbolic graph learn ing for algo- rithmic trading, in: Proceedings of the Web Conference 2021 , pp

    Sawhney, R., Agarwal, S., Wadhwa, A., Shah, R., 2021. Ex ploring the scale-free nature of stock markets: Hyperbolic graph learn ing for algo- rithmic trading, in: Proceedings of the Web Conference 2021 , pp. 11–22

    2021

  51. [59]

    Spatiotempo- ral hypergraph convolution network for stock movement fore casting, in: 2020 IEEE International Conference on Data Mining (ICDM), I EEE

    Sawhney, R., Agarwal, S., Wadhwa, A., Shah, R.R., 2020b . Spatiotempo- ral hypergraph convolution network for stock movement fore casting, in: 2020 IEEE International Conference on Data Mining (ICDM), I EEE. pp. 482–491

    2020

  52. [60]

    Seo, Y ., De fferrard, M., V andergheynst, P ., Bresson, X., 2018. Struc- tured sequence modeling with graph convolutional recurrent networks, in: Neural Information Processing: 25th International Confer ence, ICONIP 2018, Siem Reap, Cambodia, December 13-16, 2018, Proceedin gs...

    2018

  53. [61]

    Prediction of banking stock prices using naive bayes method, in: Journal of physics: Con ference se- ries, IOP Publishing

    Setiani, I., Tentua, M.N., Oyama, S., 2021. Prediction of banking stock prices using naive bayes method, in: Journal of physics: Con ference se- ries, IOP Publishing. p. 012059

    2021

  54. [62]

    S tock price fore- casting with deep learning: A comparative study

    Shahi, T.B., Shrestha, A., Neupane, A., Guo, W., 2020. S tock price fore- casting with deep learning: A comparative study. Mathemati cs 8, 1441

    2020

  55. [63]

    Assessing market integra tion between mint and developed economies: evidence from dynamic cointegrat ion

    Siddiqui, S., Kaur, R., 2023. Assessing market integra tion between mint and developed economies: evidence from dynamic cointegrat ion. Busi- ness Perspectives and Research , 22785337231156244

    2023

  56. [64]

    Autoregressive models

    Simon, L., Y oung, D., 2024. Autoregressive models. URL : https://online.stat.psu.edu/stat501/book/export/html/996

    2024

  57. [65]

    Comparis on of vector autoregressive (var) and vector error correction models (v ecm) for index of asean stock price, in: AIP Conference Proceedings, AIP Pu blishing

    Suharsono, A., Aziza, A., Pramesti, W., 2017. Comparis on of vector autoregressive (var) and vector error correction models (v ecm) for index of asean stock price, in: AIP Conference Proceedings, AIP Pu blishing

    2017

  58. [66]

    The relationship between stock price index and exchange rate in asian markets: A quantile regression approach

    Tsai, I.C., 2012. The relationship between stock price index and exchange rate in asian markets: A quantile regression approach. Jour nal of Interna- tional Financial Markets, Institutions and Money 22, 609–6 21

    2012

  59. [67]

    Determinants of stock market performance: V ar and vecm designs in korea and japan

    Tulcanaza Prieto, A.B., Lee, Y .H., 2019. Determinants of stock market performance: V ar and vecm designs in korea and japan. Global Business & Finance Review (GBFR) 24, 24–44

    2019

  60. [68]

    Wu, Z., Pan, S., Long, G., Jiang, J., Chang, X., Zhang, C. , 2020. Con- necting the dots: Multivariate time series forecasting wit h graph neural networks, in: Proceedings of the 26th ACM SIGKDD internatio nal con- ference on knowledge discovery & data mining, pp. 753–763

    2020

  61. [69]

    Gra ph wavenet for deep spatial-temporal graph modeling

    Wu, Z., Pan, S., Long, G., Jiang, J., Zhang, C., 2019. Gra ph wavenet for deep spatial-temporal graph modeling. arXiv preprint arXi v:1906.00121

    2019 arXiv

  62. [70]

    Y ahoo finance

    Y ahoo, 2024. Y ahoo finance. URL: https://finance.yahoo.com/

    2024

  63. [71]

    Spatial temporal grap h convolutional networks for skeleton-based action recognition, in: Proce edings of the AAAI conference on artificial intelligence

    Y an, S., Xiong, Y ., Lin, D., 2018. Spatial temporal grap h convolutional networks for skeleton-based action recognition, in: Proce edings of the AAAI conference on artificial intelligence

    2018

  64. [72]

    Spatio-temporal graph co nvolutional net- works: A deep learning framework for tra ffic forecasting

    Y u, B., Yin, H., Zhu, Z., 2017. Spatio-temporal graph co nvolutional net- works: A deep learning framework for tra ffic forecasting. arXiv preprint arXiv:1709.04875

    2017 arXiv

  65. [73]

    Time series forecasting using a hybr id arima and neural network model

    Zhang, G.P ., 2003. Time series forecasting using a hybr id arima and neural network model. Neurocomputing 50, 159–175

    2003

  66. [74]

    Stock market volatilit y spillovers in g7 and bric

    Zhang, P ., Sha, Y ., Xu, Y ., 2021. Stock market volatilit y spillovers in g7 and bric. Emerging Markets Finance and Trade 57, 2107–2119

    2021

  67. [75]

    Gman: A graph m ulti- attention network for traffic prediction, in: Proceedings of the AAAI con- ference on artificial intelligence, pp

    Zheng, C., Fan, X., Wang, C., Qi, J., 2020. Gman: A graph m ulti- attention network for traffic prediction, in: Proceedings of the AAAI con- ference on artificial intelligence, pp. 1234–1241

    2020

  68. [76]

    Modeling financial time serie s with S-PLUS

    Zivot, E., Wang, J., 2006. Modeling financial time serie s with S-PLUS. volume 2. Springer. 12

    2006

  69. [2023]

    IEEE Transactions on Knowledge and Data E ngineering 35, 10748–10761

    Multi-scale adaptive graph neural network for multiv ariate time se- ries forecasting. IEEE Transactions on Knowledge and Data E ngineering 35, 10748–10761

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.