Pith. sign in

REVIEW 4 major objections 5 minor 51 references

Enhancing Black-Litterman Portfolio via Hybrid Forecasting Model Combining Multivariate Decomposition and Noise Reduction

T0 review · 4 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read A three-stage forecaster that denoises returns, aligns frequency components, and deep-learns each one is proposed as the view generator for Black-Litterman portfolios, with reported annualized returns above 80 percent and Sharpe ratios…

desk verdict A useful hybrid pipeline for Black-Litterman views, but the missing walk-forward preprocessing protocol undermines the headline Sharpe results. read the letter →

arxiv 2505.01781 v2 pith:GIBA7JCB submitted 2025-05-03 cs.CE cs.AI

Ziye Yang , Ke Lu , Yang Wang , Jerome Yen This is my paper · ORCID
classification cs.CEcs.AI
keywords Black-LittermanmodelportfoliooptimizationassetpricepredictiondecompositionensembletemporalconvolutionalnetworksingularspectrumanalysisMA-EMDnoisereduction
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

The paper tries to establish that the Black-Litterman model's investor views can be generated by a three-stage forecaster: singular spectrum analysis (SSA) to remove noise, multivariate aligned empirical mode decomposition (MA-EMD) to split several price-related series into matched frequency components, and a temporal convolutional network (TCN) to predict each component one day ahead. On eight Nasdaq 100 stocks it reports that this pipeline beats three baselines on RMSE, MAPE, and $R^2$, with the largest gains on the noisy high-frequency components. It then claims that a Black-Litterman portfolio built on these forecasts outperforms mean-variance, equal-weight, and market-weight portfolios, reaching an annualized return above 80 percent and a Sharpe ratio around 4.2 under one-day rebalancing after 0.2 percent transaction costs. If the claim holds, it offers a concrete recipe for turning noisy multivariate price data into active portfolio views.

What carries the argument

The carrying mechanism is the SSA-MAEMD-TCN pipeline. SSA embeds each price series in a delay matrix, applies singular value decomposition, and reconstructs a denoised version by averaging along anti-diagonals. MA-EMD then decomposes each denoised variable separately with standard EMD and aligns the resulting intrinsic mode functions (IMFs) to the target stock's IMFs using Kullback-Leibler divergence between the distributions of extreme-value intervals; a threshold of 20 extreme values stops the decomposition. The TCN, a sequence model built from causal and dilated convolutions with residual blocks, maps seven-day windows of aligned components to the next day's value for each IMF, and the component predictions are summed and converted into returns. Those returns enter the Black-Litterman posterior through the view matrix, the view return vector, and the view error covariance matrix, and the posterior mean and covariance then drive mean-variance weights. The alignment module's key work is keeping the number and frequency content of IMFs comparable across variables, which is what lets a multivariate forecaster use volume and price-range information at matched scales.

What would settle it

Re-run the Section 4.2 pipeline with the SSA reconstruction, MA-EMD decomposition, and mean-and-standard-deviation scaling all computed inside each rolling training window only, then compare the RMSE, MAPE, $R^2$ values, and the one-day-rebalancing Sharpe ratio with Tables 4 and 8; if the Sharpe ratio falls well below 4.2 or the gaps over the baselines shrink, the central claim fails.

Watch

Extended reading notes

Core claim

On its own terms, the paper's central claim is that the SSA-MAEMD-TCN model produces markedly better next-day price forecasts for stocks than MAEMD-TCN, MEMD-TCN, and MAEMD-LSTM, and that using those forecasts as Black-Litterman views yields portfolios whose annualized returns and Sharpe ratios far exceed traditional benchmarks over short holding periods. The reported numbers include RMSE reductions on the first high-frequency component of 58 to 76 percent, full-prediction $R^2$ values mostly above 0.98, and a one-day-rebalanced Black-Litterman portfolio with cumulative return 0.5414, annual return 1.0335, and Sharpe ratio 4.2154 in the February-to-August 2023 backtest. The paper attributes these gains to SSA removing noise before decomposition and to MA-EMD aligning intrinsic mode functions across open, high, low, close, and volume series so the TCN receives consistent multivariate inputs at matched frequency scales.

Load-bearing premise

The load-bearing premise is that every preprocessing step in Section 4.2, including SSA denoising, MA-EMD decomposition, and scaling each feature by its mean and standard deviation, uses only training-window data and never data from the test or backtest period; the paper describes these steps without stating that restriction, so if full-sample preprocessing was used the reported out-of-sample forecasts and portfolio gains would be contaminated by future information.

Editorial extensions

If this is right

  • If the central claim is right, investors can replace subjective or single-variable views in Black-Litterman with views from a denoised, frequency-aligned multivariate forecast pipeline.
  • The noise-reduction stage is doing substantial work: SSA-MAEMD-TCN beats MAEMD-TCN on every tested stock, and the largest per-component gains appear on the first IMF, where noise concentrates.
  • MA-EMD outperforms MEMD inside the same TCN forecaster for most stocks, supporting the paper's argument that aligned decomposition quality matters more than joint multivariate decomposition.
  • The Black-Litterman portfolio is more diversified than mean-variance optimization, with an average of about 18 stocks and a lower Herfindahl-Hirschman Index, reducing concentration risk under daily rebalancing.
  • The advantage decays with holding period: at 20-day rebalancing the Black-Litterman Sharpe ratio falls toward 3.39 and the edge over market-weight and equal-weight benchmarks narrows, so the claim is about short-horizon, high-frequency allocation.

Reading between the lines

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

  • The sequence learner is probably not the only source of gain: rerunning the same SSA-MAEMD pipeline with LSTM, GRU, or a transformer would show whether the reported edge is tied to TCN or to the denoising and alignment stages.
  • The alignment module is not finance-specific; the same Kullback-Leibler divergence alignment on extreme-value intervals could be applied to any multi-channel signal whose channels oscillate at different scales, such as sensor arrays or physiological recordings.
  • The paper itself notes in Section 6 that the model only performs single-step daily forecasts, so the strong results should be read as evidence about short-horizon tactical allocation rather than weekly or monthly holding periods, which would need a multi-step forecaster.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes a hybrid stock-price forecasting pipeline, SSA-MAEMD-TCN, that combines singular spectrum analysis for denoising, multivariate aligned empirical mode decomposition, and temporal convolutional networks. The forecasts are used as investor views in a Black-Litterman portfolio model. Using daily data for eight Nasdaq-100 stocks from January 2020 to August 2023, the authors report that the proposed model improves RMSE, MAPE, and R-squared relative to MAEMD-TCN, MEMD-TCN, and MAEMD-LSTM, and that a Black-Litterman portfolio built on these views achieves annualized returns above 80% and Sharpe ratios near 4.2 with 1-day rebalancing after transaction costs. The paper includes detailed tables of forecast metrics, portfolio performance, and diversification measures.

Significance. If the preprocessing were verified to be leakage-free, the combination of SSA denoising, MA-EMD, and TCN for Black-Litterman view generation would be a useful empirical contribution, and the paper provides complete numerical tables of forecast and portfolio metrics. However, the central empirical claim depends on the out-of-sample validity of the pipeline, and the manuscript does not document that normalization, SSA, or EMD are computed in a walk-forward manner. The paper also provides no code, no error bars, and no significance tests, so the reported improvements and Sharpe ratios cannot currently be distinguished from artifacts of lookahead or estimation noise. The contribution is therefore plausible but not yet established.

major comments (4)
  1. [Section 4.2, Eq. (22); Sections 3.4 and 3.2.1] The paper never states that normalization, SSA denoising, and MA-EMD decomposition are fit only on training data. Eq. (22) defines z-scores with global mu_i and sigma_i, Section 3.4 constructs the SSA trajectory matrix over the full series and applies SVD, and Section 3.2.1 sifts EMD envelopes over the full signal. If these steps are applied before the 70/15/15 split, the test forecasts in Table 4 and the 1-day-rebalancing Sharpe ratio of 4.2154 in Table 8 are contaminated by future information. Please specify an explicit leakage-free protocol: fit mu/sigma, SSA basis, and EMD envelopes on each training window only, apply them to subsequent data, and re-estimate them at each rebalancing or rolling date. Providing pseudocode or code for this protocol would be essential.
  2. [Section 5.1, Table 4] The claim that SSA-MAEMD-TCN is "significantly better" than the benchmarks rests on point estimates from a single train/validation/test split with no error bars, no multiple random seeds, and no significance tests. TCN training is stochastic, and the test period is one 132-day window, so the reported differences could be within noise. Please report mean and standard deviation over at least five seeds, include paired Diebold-Mariano or bootstrap tests for RMSE/MAPE differences, and justify the selection of 8 of the 20 stocks in Table 1. Without this, the word "significant" is unsupported.
  3. [Section 5.2, Tables 7 and 8] The backtest covers 132 trading days with daily rebalancing, and the paper reports a Sharpe ratio of 4.2154 net of 0.2% transaction costs. The paper does not report turnover, gross return, or cost sensitivity. For an 18-stock portfolio rebalanced daily, 0.2% per trade is material, and the cost assumption heavily affects the comparison with the equal-weight and market-weight benchmarks. Please report gross and net annualized returns, average one-way and round-trip turnover, average daily cost as a fraction of portfolio value, and a sensitivity analysis over transaction cost levels.
  4. [Section 5.2.1] The rolling data scheme is described as constructing a portfolio on the initial date, holding it for a fixed period, and then reconstructing "using forecast data from the second day of the test period." This is ambiguous: it is not clear whether investor views are regenerated daily from the TCN, whether the forecast horizon matches the holding period, or how the 1-day, 3-day, 5-day, 10-day, and 20-day horizons are handled. Please clarify the exact calendar of re-estimation and rebalancing, and state whether the same one-step-ahead forecast is used for all holding periods. This ambiguity, together with the preprocessing concern, makes the short-horizon advantage difficult to interpret.
minor comments (5)
  1. [Title] The title says "Empirical Modal Decomposition" but the correct term used throughout the body is "Empirical Mode Decomposition." Please correct the title.
  2. [Section 4.1, Tables 1 and 2] Table 1 lists 20 stocks, but Table 2 reports statistics for only 8 stocks. Please state explicitly whether the other 12 stocks are used in the portfolio construction or only in the subset of forecasts.
  3. [Section 3.1.2, Eq. (2)] The view matrix P is displayed with the entry 1 in the same column in every row; for k absolute views on different assets, the 1s should appear in different columns according to the asset of each view. Please fix the display or clarify that the illustration is schematic.
  4. [Section 4.2, Eq. (22)] The text says "Before decomposition, normalizing the data is not negligible" but does not clarify whether normalization precedes SSA denoising or follows it. Please state the order of all preprocessing steps explicitly.
  5. [Section 5.1, Table 5] The table and text refer to "GOOGLE" when the stock identifier is GOOGL. Also, the text says R^2 ranges from 0 to 1, but Table 5 reports negative R^2 values for some benchmark fits; please reconcile this description.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the forecasting and Black-Litterman results are empirical outputs, not reductions of their inputs by construction.

full rationale

The paper is an empirical pipeline rather than a derivation, and I find no step in which a claimed prediction is equivalent to its input by definition or by self-citation. The views Q in Eq. (3) are produced by the SSA-MAEMD-TCN forecasts, while P in Eq. (2) is a fixed selection matrix and the posterior return in Eq. (5) is the standard Black-Litterman Bayesian update; nothing in these equations makes the output equal to a fitted constant. The forecasting comparison in Table 4 evaluates RMSE, MAPE, and R2 against independent benchmark models, so the claimed improvement is an empirical outcome, not a constructional identity. The MA-EMD component is credited to Cai et al. (2025), an external published source, and the paper does not invoke a uniqueness theorem or rely on a load-bearing self-citation chain; none of the authors of the present work authored the cited MA-EMD algorithm paper. The one substantive concern raised by the text is potential lookahead: Eq. (22) and the SSA and MA-EMD preprocessing in Section 4.2 are described before the 70/15/15 split is stated, and global mu_i, sigma_i, SVD bases, and EMD envelopes could in principle be computed on the full sample. That is a data-leakage and correctness issue, not a circularity issue, because even if the preprocessing used future information, the reported forecasts and Sharpe ratios would be inflated empirical estimates rather than results that reduce to their own inputs by definition. No equation in the paper makes the outcome equal to the input, and no fitted parameter is renamed as a prediction. Accordingly, the honest finding is no significant circularity, score 0.

Assumptions & free parameters 7 free parameters · 6 assumptions · 0 invented entities

No new theoretical entities are introduced. The central claim rests on the Black-Litterman equilibrium-prior assumptions, the validity of EMD and SSA as signal separators, and a clean train-test split. The last of these is not documented and is the main risk to the empirical claims.

free parameters (7)
  • BL risk aversion lambda = 2.5
    Set in Section 3.1.1 as the world average risk aversion index; affects all portfolio weights.
  • BL scalar tau = 1/500
    Set in Section 3.1.2 based on 500 days of historical data; scales view uncertainty.
  • MA-EMD extreme-value threshold omega = 20
    Stops the EMD decomposition process; chosen in Section 3.2.2 with no sensitivity analysis.
  • SSA window length and number of retained components = not specified
    Required for SSA denoising; not disclosed in Section 4.2, which blocks replication and leaves the noise-signal split undefined.
  • TCN sliding window length = 7
    Seven past days are used to predict the eighth data point; chosen in Section 4.2.
  • TCN hyperparameters = kernel 2, hidden [64,128], lr 0.001, layers 2, dropout [0,0.3], epochs [50,100], batch [16,32]
    Listed in Table 3; the tuning procedure and whether only the validation set guided the choices are not described.
  • Transaction cost = 0.2%
    Assumed in the rebalancing backtest in Section 5.2.2; no sensitivity analysis around this value.
assumptions (6)
  • domain assumption EMD produces IMFs that are meaningful predictive components for financial price series.
    The entire M-DEM pipeline assumes IMFs capture distinct time-scale patterns; no theoretical guarantee for financial data; used in Section 3.2.1.
  • domain assumption SSA denoising removes noise while preserving the predictive signal.
    SSA grouping parameters are not specified; effectiveness is only indirectly tested; Sections 3.4 and 4.2.
  • domain assumption Market equilibrium returns are a valid Bayesian prior for expected returns.
    The Black-Litterman reverse optimization in Section 3.1.1 assumes the market portfolio is mean-variance efficient.
  • domain assumption Past price patterns generalize from the training window to the test window.
    The train, validation, and test split in Section 4.2 assumes enough stationarity for TCN forecasts to transfer.
  • domain assumption KLD-based frequency alignment preserves information needed for forecasting.
    MA-EMD aligns IMFs by extremal-interval distributions, a heuristic from Cai et al.; Section 3.2.2.
  • ad hoc to paper Preprocessing statistics are computed without future information.
    The paper does not state that SSA and normalization in Eq. (22) are fit only on training windows; this undocumented assumption is load-bearing for the out-of-sample claim.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing Black-Litterman Portfolio via Hybrid Forecasting Model Combining Multivariate Decomposition and Noise Reduction." pith.science (2026). https://pith.science/paper/GIBA7JCB

@misc{pith2026250501781,
  author       = {Pith},
  title        = {Pith review of: Enhancing Black-Litterman Portfolio via Hybrid Forecasting Model Combining Multivariate Decomposition and Noise Reduction},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/GIBA7JCB}},
  note         = {Machine review of arXiv:2505.01781}
}
read the original abstract

Modern portfolio construction demands robust methods for integrating data-driven insights into asset allocation. The Black-Litterman model offers a powerful Bayesian approach to adjust equilibrium returns using investor views to form a posterior expectation along with market priors. Mainstream research mainly generates subjective views through statistical models or machine learning methods, among which hybrid models combined with decomposition algorithms perform well. However, most hybrid models do not pay enough attention to noise, and time series decomposition methods based on single variables make it difficult to fully utilize information between multiple variables. Multivariate decomposition also has problems of low efficiency and poor component quality. In this study, we propose a novel hybrid forecasting model SSA-MAEMD-TCN to automate and improve the view generation process. The proposed model combines Singular Spectrum Analysis (SSA) for denoising, Multivariate Aligned Empirical Mode Decomposition (MA-EMD) for frequency-aligned decomposition, and Temporal Convolutional Networks (TCNs) for deep sequence learning to capture complex temporal patterns across multiple financial indicators. Empirical tests on the Nasdaq 100 Index stocks show a significant improvement in forecasting performance compared to baseline models based on MAEMD and MEMD. The optimized portfolio performs well, with annualized returns and Sharpe ratios far exceeding those of the traditional portfolio over a short holding period, even after accounting for transaction costs.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

51 extracted references · 31 canonical work pages

  1. [1]

    Ak¸ sehir, Z.D., & Kılı¸ c, E. (2024). A new denoising approach based on mode decom- position applied to the stock market time series: 2le-ceemdan. PeerJ Computer Sci- ence, 10, e1852, https://doi.org/10.7717/ peerj-cs.1852

    work page 2024

  2. [2]

    Alrumaih, R.M., & Al-Fawzan, M.A. (2002). Time series forecasting using wavelet denoising an application to saudi stock index. Journal of King Saud University-Engineering Sciences, 14(2), 221–233, https://doi.org/10.1016/ S1018-3639(18)30755-4

    work page 2002

  3. [3]

    Bai, J., Guo, J., Sun, B., Guo, Y., Bao, Q., Xiao, X. (2023). Intelligent forecasting model of stock price using neighborhood rough set and multivariate empirical mode decomposi- tion. Engineering Applications of Artificial Intelligence, 122, 106106, https://doi.org/ 10.1016/j.engappai.2023.106106

    arXiv 2023

  4. [4]

    (2018, April)

    Bai, S., Kolter, J.Z., Koltun, V. (2018, April). An Empirical Evaluation of Generic Convolu- tional and Recurrent Networks for Sequence Modeling (No. arXiv:1803.01271). arXiv

    arXiv 2018

  5. [5]

    Bao, W., Yue, J., Rao, Y. (2017). A deep learning framework for financial time series using stacked autoencoders and long- short term memory. PloS one , 12(7), e0180944, https://doi.org/10.1371/journal .pone.0180944

    doi:10.1371/journal 2017

  6. [6]

    Barua, R., & Sharma, A.K. (2022). Dynamic black litterman portfolios with views derived via cnn-bilstm predictions. Finance Research Letters, 49, 103111, https://doi.org/10 .1016/j.frl.2022.103111 18

    arXiv 2022

  7. [7]

    Barua, R., & Sharma, A.K. (2023). Using fear, greed and machine learning for opti- mizing global portfolios: A black-litterman approach. Finance Research Letters , 58, 104515, https://doi.org/10.1016/j.frl.2023 .104515

    work page doi:10.1016/j.frl.2023 2023

  8. [8]

    Beach, S.L., & Orlov, A.G. (2007). An application of the black–litterman model with egarch- m-derived views for international portfolio management. Financial Markets and Port- folio Management , 21, 147–166, https:// doi.org/10.1007/s11408-007-0046-6

    work page doi:10.1007/s11408-007-0046-6 2007

Show all 51 references
  1. [9]

    Best, M.J., & Grauer, R.R. (1991). On the sen- sitivity of mean-variance-efficient portfolios to changes in asset means: some analytical and computational results. The review of financial studies , 4(2), 315–342, https:// doi.org/10.1093/rfs/4.2.315

    1991 doi

  2. [10]

    Black, F. (1986). Noise. The journal of finance , 41(3), 528–543, https://doi.org/10.1111/ j.1540-6261.1986.tb04513.x

    1986

  3. [11]

    Black, F., & Litterman, R. (1992). Global portfo- lio optimization. Financial analysts journal , 48(5), 28–43, https://doi.org/10.2469/faj .v48.n5.28

    1992 doi

  4. [12]

    Broomhead, D.S., & King, G.P. (1986). Extract- ing qualitative dynamics from experimental data. Physica D: Nonlinear Phenomena , 20(2-3), 217–236, https://doi.org/10.1016/ 0167-2789(86)90031-X

    1986

  5. [13]

    Cai, X., & Li, D. (2024). M-edem: A mnn-based empirical decomposition ensem- ble method for improved time series fore- casting. Knowledge-Based Systems , 283, 111157, https://doi.org/10.1016/j.knosys .2023.111157

    2024

  6. [14]

    Cai, X., Li, D., Zhang, J., Wu, Z. (2025). Ma-emd: Aligned empirical decomposi- tion for multivariate time-series forecast- ing. Expert Systems with Applications , 267, 126080, https://doi.org/10.1016/j .eswa.2024.126080

    2025

  7. [15]

    Deng, C., Huang, Y., Hasan, N., Bao, Y. (2022). Multi-step-ahead stock price index forecast- ing using long short-term memory model with multivariate empirical mode decompo- sition. Information Sciences , 607, 297–321, https://doi.org/10.1016/j.ins.2022.05.088

    2022 doi

  8. [16]

    Duqi, A., Franci, L., Torluccio, G. (2014). The black–litterman model: the defini- tion of views based on volatility forecasts. Applied Financial Economics, 24(19), 1285– 1296, https://doi.org/10.1080/09603107 .2014.925056

    2014

  9. [17]

    Gao, P., Zhang, R., Yang, X. (2020). The application of stock index price prediction with neural network. Mathematical and Computational Applications , 25(3), 53,

    2020

  10. [18]

    Hassani, H., Dionisio, A., Ghodsi, M. (2010). The effect of noise reduction in measuring the linear and nonlinear dependency of financial markets. Nonlinear Analysis: Real World Applications, 11(1), 492–502, https://doi .org/10.1016/j.nonrwa.2009.01.004

    2010 doi

  11. [19]

    He, G., & Litterman, R. (2002). The intu- ition behind black-litterman model portfo- lios. Available at SSRN 334304 , , https:// doi.org/10.2139/ssrn.334304

    2002 doi

  12. [20]

    Liu, H.H

    Shih, H.H., Zheng, Q., . . . Liu, H.H. (1998). The empirical mode decomposition and the hilbert spectrum for nonlinear and non- stationary time series analysis. Proceedings of the Royal Society of London. Series A: mathematical, physical and engineering sci- ences, 454(1971), ...

    1998

  13. [21]

    Idzorek, T. (2007). A step-by-step guide to the black-litterman model: Incorporating user-specified confidence levels. Forecast- ing expected returns in the financial markets (pp. 17–38). Elsevier

    2007

  14. [22]

    Jiang, W. (2021). Applications of deep learning in stock market prediction: recent progress. Expert Systems with Applications , 184, 115537, https://doi.org/10.1016/j .eswa.2021.115537

    2021

  15. [23]

    Kara, M., Ulucan, A., Atici, K.B. (2019). A hybrid approach for generating investor views in black–litterman model. Expert Systems with Applications, 128, 256–270, https://doi .org/10.1016/j.eswa.2019.03.041

    2019 doi

  16. [24]

    Kumbure, M.M., Lohrmann, C., Luukka, P., Por- ras, J. (2022). Machine learning techniques and data for stock market forecasting: A lit- erature review. Expert Systems with Appli- cations, 197, 116659, https://doi.org/10 .1016/j.eswa.2022.116659

    2022

  17. [25]

    Lahmiri, S. (2018). Minute-ahead stock price forecasting based on singular spectrum anal- ysis and support vector regression. Applied Mathematics and Computation , 320, 444– 451, https://doi.org/10.1016/j.amc.2017.09 .049

    2018 doi

  18. [26]

    Horch, A., Su, H. (2018). Fast multivari- ate empirical mode decomposition. IEEE Access, 6, 65521–65538, https://doi.org/ 10.1109/ACCESS.2018.2877150

    2018

  19. [27]

    Lei, D. (2019). Black–litterman asset alloca- tion model based on principal component analysis (pca) under uncertainty. Cluster Computing, 22, 4299–4306, https://doi .org/10.1007/s10586-018-1864-1

    2019 doi

  20. [28]

    Lin, Y., Yan, Y., Xu, J., Liao, Y., Ma, F. (2021). Forecasting stock index price using the ceemdan-lstm model. The North Amer- ican Journal of Economics and Finance , 57, 101421, https://doi.org/10.1016/j.najef .2021.101421

    2021

  21. [29]

    Markowitz, H. (1952). Portfolio selection. The Journal of Finance , 7(1), 77–91, https:// doi.org/10.2307/2975974

    1952 doi

  22. [30]

    Michaud, R.O. (1989). The markowitz optimiza- tion enigma: Is ‘optimized’optimal? Finan- cial analysts journal , 45(1), 31–42, https:// doi.org/10.2139/ssrn.2387669

    1989 doi

  23. [31]

    Mourad, N. (2023). Multivariate group-sparse mode decomposition. Digital Signal Process- ing, 137, 104024, https://doi.org/10.1016/ j.dsp.2023.104024

    2023

  24. [32]

    Palomba, G. (2008). Multivariate garch models and the black-litterman approach for track- ing error constrained portfolios: an empirical analysis. Global Business and Economics Review, 10(4), 379–413, https://doi.org/ 10.1504/GBER.2008.020592

    2008 arXiv

  25. [33]

    Pyo, S., & Lee, J. (2018). Exploiting the low-risk anomaly using machine learning to enhance the black–litterman framework: Evidence from south korea. Pacific-Basin Finance Journal , 51, 1–12, https://doi .org/10.1016/j.pacfin.2018.06.002

    2018 doi

  26. [34]

    Rehman, N., & Mandic, D.P. (2010). Multivariate empirical mode decomposition. Proceedings of the Royal Society A: Mathematical, Phys- ical and Engineering Sciences , 466(2117), 1291–1302, https://doi.org/10.1098/rspa .2009.0502

    2010

  27. [35]

    Rezaei, H., Faaljou, H., Mansourfar, G. (2021). Intelligent asset allocation using predic- tions of deep frequency decomposition. 20 Expert Systems with Applications , 186, 115715, https://doi.org/10.1016/j.eswa .2021.115715

    2021

  28. [36]

    Rilling, G., Flandrin, P., Gon¸ calves, P., Lilly, J.M. (2007). Bivariate empirical mode decom- position. IEEE signal processing letters , 14(12), 936–939, https://doi.org/10.1109/ LSP.2007.904710

    2007

  29. [37]

    Satchell, S., & Scowcroft, A. (2007). A demysti- fication of the black-litterman model: Man- aging quantitative and traditional portfolio construction. Forecasting expected returns in the financial markets (pp. 39–53). Elsevier

    2007

  30. [38]

    Shu, W., & Gao, Q. (2020). Forecasting stock price based on frequency components by emd and neural networks. IEEE Access, 8, 206388–206395, https://doi.org/10.1109/ ACCESS.2020.3037681

    2020

  31. [39]

    Tang, Q., Fan, T., Shi, R., Huang, J., Ma, Y. (2021). Prediction of financial time series using lstm and data denoising methods

    2021

  32. [40]

    Ji, J., . . . Li, J. (2022). A survey on machine learning models for financial time series forecasting. Neurocomputing, 512, 363-380, https://doi.org/10.1016/j.neucom .2022.09.003

    2022 doi

  33. [41]

    Thirumalaisamy, M.R., & Ansell, P.J. (2018). Fast and adaptive empirical mode decomposition for multidimensional, multivariate signals. IEEE Signal Processing Letters , 25(10), 1550–1554, https://doi.org/10.1109/LSP .2018.2867335

    2018

  34. [42]

    Flandrin, P. (2011). A complete ensemble empirical mode decomposition with adap- tive noise. 2011 ieee international confer- ence on acoustics, speech and signal process- ing (icassp) (pp. 4144–4147). ur Rehman, N., & Mandic, D.P. (2009). Empirical mode decomposition for triva...

    2011

  35. [43]

    Wang, C.-H., Yuan, J., Zeng, Y., Lin, S. (2024). A deep learning integrated framework for predicting stock index price and fluctuation via singular spectrum analysis and particle swarm optimization. Applied Intelligence , 54(2), 1770–1797, https://doi.org/10.1007/ s10489-024-05271-x

    2024

  36. [44]

    Wang, J., & Liu, J. (2024). Two-stage deep ensem- ble paradigm based on optimal multi-scale decomposition and multi-factor analysis for stock price prediction. Cognitive Compu- tation, 16(1), 243–264, https://doi.org/ 10.1007/s12559-023-10203-x

    2024 doi

  37. [45]

    Wu, Z., & Huang, N.E. (2009). Ensemble empir- ical mode decomposition: a noise-assisted data analysis method. Advances in adaptive data analysis, 1(01), 1–41, https://doi.org/ 10.1142/S1793536909000047

    2009 doi

  38. [46]

    Yao, Y., yang Zhang, Z., Zhao, Y. (2023). Stock index forecasting based on mul- tivariate empirical mode decomposition and temporal convolutional networks. Applied Soft Computing , 142, 110356, https://doi.org/https://doi.org/10.1016/ j.asoc.2023.110356

    2023

  39. [47]

    Yeh, J.-R., Shieh, J.-S., Huang, N.E. (2010). Complementary ensemble empirical mode decomposition: A novel noise enhanced data analysis method. Advances in adaptive data analysis, 2(02), 135–156, https://doi.org/ 10.1142/S1793536910000422

    2010 doi

  40. [48]

    Yu, H., Ming, L.J., Sumei, R., Shuping, Z. (2020). A hybrid model for financial time series fore- casting—integration of ewt, arima with the improved abc optimized elm. IEEE Access, 21 8, 84501–84518, https://doi.org/10.1109/ ACCESS.2020.2987547

    2020

  41. [49]

    Zhang, C., Sjarif, N.N.A., Ibrahim, R. (2024). Deep learning models for price forecasting of financial time series: A review of recent advancements: 2020–2022. Wiley Interdisci- plinary Reviews: Data Mining and Knowl- edge Discovery, 14(1), e1519,

    2024

  42. [50]

    Zhang, Y., Yan, B., Aasma, M. (2020). A novel deep learning framework: Prediction and analysis of financial time series using ceemd and lstm. Expert systems with appli- cations, 159, 113609, https://doi.org/10 .1016/j.eswa.2020.113609

    2020

  43. [51]

    Zou, Y., & He, K. (2022). Forecasting crude oil risk using a multivariate multiscale convolu- tional neural network model. Mathematics, 10(14), 2413, https://doi.org/10.3390/ math10142413 22

    2022

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.