Pith. sign in

REVIEW 5 major objections 5 minor 47 references

Dynamic ETF Portfolio Optimization Using enhanced Transformer-Based Models for Covariance and Semi-Covariance Prediction(Work in Progress)

T0 review · 5 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read The paper claims that Transformer forecasts of covariance and semi-covariance matrices outperform the historical sample as inputs to minimum-variance ETF portfolios, with semi-covariance generally improving downside-risk-adjusted…

desk verdict A plausible covariance-forecasting pipeline undermined by a one-month backtest, an undisclosed selection cutoff that may leak the test window, and internally inconsistent reported numbers. read the letter →

arxiv 2411.19649 v1 pith:C536TU5P submitted 2024-11-29 q-fin.PM q-fin.CP

Jiahao Zhu , Hengzhi Wu This is my paper · ORCID
classification q-fin.PMq-fin.CP MSC 91G1091G8068T07
keywords FinancialForecastingDeepLearningTransformerCovarianceSemi-CovariancePortfolioOptimizationETFSortinoratio
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 claims that Transformer-based models—Autoformer, Informer, Reformer, and the vanilla Transformer—can forecast the full covariance and semi-covariance matrices of a diversified ETF pool more accurately than the sample historical method, and that plugging these forecasts into minimum-variance optimization improves portfolio performance. The central selling point is the semi-covariance matrix, which measures only downside co-movement and better matches risk-averse investors' concern with losses; the paper reports that portfolios optimized on predicted semi-covariance generally beat those optimized on predicted covariance in return and Sortino ratio. A sympathetic reader would care because this is a concrete template for replacing static risk estimates with a data-driven, attention-based forecaster while keeping the optimization layer unchanged. The paper presents this as a working version with a single one-month test window, so the reported magnitudes are provisional.

What carries the argument

The central machinery is a compression–Transformer–reconstruction pipeline. The unique entries of each day's covariance or semi-covariance matrix are flattened into a vector, fed as a time series into an attention-based sequence model (Autoformer, Informer, Reformer, or vanilla Transformer), and the forecast vector is reshaped back into a matrix. Symmetry and positive semi-definiteness are enforced by a regularization loss (penalizing asymmetry and negative eigenvalues) and by a post-hoc eigenvalue-clipping projection to the nearest PSD matrix. The resulting matrix is then substituted directly into the closed-form minimum-variance weight formula $w = \frac{\Sigma^{-1}\mathbf{1}}{\mathbf{1}^{\top}\Sigma^{-1}\mathbf{1}}$, so the only change relative to the baseline is the source of $\Sigma$.

What would settle it

A concrete check is to re-run the experiment with the ETF pool fixed using only data up to February 11, 2024 (the day before the test set starts), or with a randomly selected pool, and to extend the test to several non-overlapping months; if the MSE and Sortino improvements for the Transformer and semi-covariance approaches vanish under this out-of-sample selection, the reported edge is a look-ahead artifact rather than a forecasting gain.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that the attention mechanism can learn time-varying cross-asset dependence directly from the compressed triangular part of a covariance matrix, and that the resulting forecasts are better inputs to minimum-variance optimization than the trailing historical matrix. The paper further claims that the semi-covariance matrix, because it isolates downside co-movement, is the more suitable risk input for risk-averse investors: across its tests, semi-covariance-based portfolios mostly produced higher returns and higher Sortino ratios than covariance-based portfolios, with the clearest gains for the vanilla Transformer and Informer. The reported numbers support the direction of the claim for prediction accuracy—all Transformer variants lowered MSE for both matrices—while the performance edge varies by model, and the paper reads the overall pattern as evidence that downside-focused dynamic forecasts adapt better to volatile conditions.

Load-bearing premise

The load-bearing premise is that the ETF selection rule—choosing, within each asset class, the funds with the highest three-year historical returns—does not leak information about the test month, which falls inside that same three-year window; if the top pick was already a winner during the test month, the backtest advantage could come from selection rather than from Transformer forecasting.

Editorial extensions

If this is right

  • Asset managers could adopt the pipeline as a drop-in replacement for the covariance input in minimum-variance optimization, without changing the optimizer.
  • If the semi-covariance effect holds, risk-averse portfolios can be built to target downside co-movement directly instead of total volatility.
  • The compression–reconstruction trick makes Transformer forecasting of full covariance matrices feasible even when the number of assets makes the raw matrix high-dimensional.
  • The PSD-enforcement scheme offers a practical solution to the non-positive-definiteness problem that has historically blocked semi-covariance optimization.

Reading between the lines

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

  • The reported test is a single month with undisclosed rolling-window settings, so the specific return and Sortino numbers are likely optimistic; a multi-year out-of-sample backtest would be needed to confirm the edge is stable.
  • Because the ETF selection rule looks at three-year returns ending after the test month, the fairest reading is that this paper demonstrates a promising methodology, not a validated investment edge.
  • The mixed results for Autoformer and Reformer—where covariance actually produced higher returns than semi-covariance—suggest the downside-risk advantage may be model-specific rather than a universal property of semi-covariance optimization.
  • The same compression–transformer template could be applied to other non-PSD risk measures, such as expected-shortfall contributions or higher-moment co-skewness matrices.
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

5 major / 5 minor

Summary. The manuscript proposes an end-to-end pipeline in which the lower-triangular entries of rolling covariance and semi-covariance matrices of an ETF pool are compressed into vectors, fed into Transformer variants (vanilla Transformer, Autoformer, Informer, Reformer), and reconstructed as forecast matrices, with regularization and eigenvalue clipping used to enforce positive semi-definiteness. The forecasts are then used to construct minimum-variance portfolios. The paper reports MSE improvements in Table 1 and return/Sortino improvements in Table 2 relative to a sample-covariance baseline, and concludes that Transformer-based dynamic risk-matrix forecasts, especially with the semi-covariance matrix, significantly improve portfolio performance.

Significance. Forecasting a semi-covariance matrix with attention-based models and using it directly in minimum-variance optimization is a reasonable research direction, and the paper is careful to motivate the downside-risk objective and to compare against a standard baseline. The MSE table (Table 1) provides a concrete quantitative target, and the use of the Sortino ratio is appropriate given the downside-risk focus. However, the empirical evidence is not sufficient to support the abstract's strong claims: the test period is a single month, the ETF selection rule may overlap the test window, and key rolling-window parameters are withheld. As it stands, the contribution is an architectural proposal with an illustrative backtest rather than a validated result.

major comments (5)
  1. [Section 4.1 (Data Description and Experimental Setup)] The asset-selection rule chooses, within each asset class, the ETF with the highest expected return over the past three years, but the text never states the cutoff date for that three-year lookback. The observed data are stated to span March 21, 2022 to March 21, 2024, and the test set is February 12 to March 12, 2024. If the ranking is computed as of any date on or after February 12, 2024, the test month is inside the selection window, so the ETF pool in Table 3 may be selected using returns from the test period. Because the stated data span is only two years, the three-year history must come from an unstated external source, so the cutoff cannot be inferred. This potential lookahead bias undermines the out-of-sample interpretation of Table 2 and Figures 2 and 3.
  2. [Section 5.1 and Table 2] The entire performance comparison rests on a single one-month test window (February 12 to March 12, 2024). No standard errors, confidence intervals, or significance tests are reported for the return or Sortino differences in Table 2, and the MSE values in Table 1 are given as averages over five runs without dispersion. The abstract's claim that the predictions 'significantly enhance portfolio performance' is therefore not supported by the evidence presented; the observed differences could easily be within noise for one month and a small pool of assets.
  3. [Section 4.1 (rolling window)] The paper states that the precise rolling-window parameters, including window size and rebalancing frequency, 'are not disclosed here due to confidentiality restrictions.' Without these parameters, the experiment cannot be reproduced, the number of effective rebalancing periods is unknown, and it is unclear whether the MSE and return results are based on overlapping or independent predictions. This lack of disclosure is a load-bearing gap for any empirical claim in the paper.
  4. [Section 5.4 and Table 2] The Informer results are internally inconsistent. Table 2 reports Return using Covariance = 1.17% and Sortino using Covariance = 3.31, but Section 5.4 states that Informer's 'return increase from 3,31% to 5.69%, and its Sortino ratio improves from 3.31% to 10.05.' The text appears to conflate the return and the Sortino ratio and contradicts the table, making the numerical results unreliable for at least one model.
  5. [Sections 3.4-3.5] The post-processing step clips negative eigenvalues to zero (Section 3.5) to enforce positive semi-definiteness, while the regularized loss in Section 3.4 penalizes negative eigenvalues only softly. For the semi-covariance matrix, which is not positive semi-definite by construction (Section 2.4), this projection can materially alter the risk measure before portfolio optimization. The paper does not quantify the distance between the projected and unprojected matrices or test whether the reported portfolio improvements survive without the projection.
minor comments (5)
  1. [Section 4.1] The data window contains a typo: the text first states data span from March 21, 2022 to March 21, 2024, but the next sentence gives the training set as March 12, 2022 to February 12, 2024; the training-set start date should presumably be March 21, 2022.
  2. [Section 4.2] The bullet labeled 'Actual Result' is unclear; it is not a model and appears to be a leftover label rather than a description of a baseline.
  3. [Section 5.4] The text uses a European comma decimal ('3,31%') inconsistently with the rest of the paper, and the bullet about Informer mixes return and Sortino units.
  4. [Data Availability Statement] The Data Availability Statement points to an 'EarningsCall Dataset' repository, which does not match the ETF data used in this study.
  5. [Section 5.2] Some sentences are incomplete or run-on, for example 'To further investigate the model's behavior.' appears without a continuation.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the forecasting pipeline is trained on observed matrices and evaluated on realized returns, with only non-circular validity caveats.

full rationale

The derivation chain is not circular. The Transformer models are trained on a mean-squared-error loss against observed covariance and semi-covariance matrices (Section 3.4) and are then evaluated on realized one-month portfolio returns and Sortino ratios (Section 4.2), an outcome variable that is not used in the training loss. The MSE comparisons in Table 1 and the portfolio comparisons in Table 2 are therefore not definitionally forced by the training objective. The main methodological caveats are the unspecified cutoff for the three-year ETF-selection lookback (Section 4.1), which creates a potential lookahead risk if the test month falls inside the ranking window, and the alignment between the semi-covariance objective and the Sortino evaluation metric (Sections 2.4 and 4.2), which can make the semi-covariance-versus-covariance comparison look favorable. These are threats to out-of-sample validity rather than reductions by construction: the Transformer-versus-sample prediction is not equivalent to its input. The authors explicitly acknowledge the short-period and concentration caveats in Section 5.2 ('these gains may be partly attributed to the greater concentration of the portfolio ... or the selected period') and list longer out-of-sample testing as future work in Section 6.4. No load-bearing self-citation chain or imported uniqueness theorem is used. Hence no significant circularity.

Assumptions & free parameters 8 free parameters · 5 assumptions · 0 invented entities

The central claims rest on hyperparameters tuned on validation MSE, on two undisclosed experimental parameters (rolling window size and rebalancing frequency), and on the assumption that nearest-PSD projection preserves the downside-risk signal. The asset-selection horizon of three years overlaps the one-month test window. No new entities are introduced.

free parameters (8)
  • learning_rate = 0.0001
    Hyperparameter tuned on validation MSE; affects convergence but not the central claim directly.
  • dropout_rate = 0.05
    Hyperparameter chosen by validation.
  • num_epochs = 100
    Hyperparameter chosen by validation.
  • d_model = 256
    Hidden dimension of the covariance reconstruction module.
  • regularization_weight_alpha
    Penalty factor in Ltotal = Lsingle + alpha(Lsym + LPSD); no value is stated in the paper.
  • rolling_window_size
    Undisclosed due to confidentiality; essential for reproducing the experiment.
  • rebalancing_frequency
    Undisclosed due to confidentiality; essential for reproducing the experiment.
  • expected_return_selection_horizon = 3 years
    Asset selection uses highest expected returns over the past three years; overlaps with the one-month test window.
assumptions (5)
  • standard math Lagrange multiplier solution for the minimum-variance portfolio w = Sigma^{-1}1 / (1^T Sigma^{-1}1).
    Used in Section 2.1 to derive portfolio weights from the predicted covariance or semi-covariance matrix; assumes Sigma is invertible.
  • domain assumption Asset correlations are relatively stable over time.
    Section 3.2 states the authors are 'convinced by the work of Bollerslev (1986) and Engle (2002)' that correlation structure is stable, which justifies using historical attention patterns to forecast future covariance.
  • domain assumption Semi-covariance, defined as E[min(X-µX,0) min(Y-µY,0)], is a better risk measure for risk-averse investors.
    Section 2.4 adopts Estrada's semi-covariance definition and claims it aligns with downside-risk preferences; this motivates the entire comparison.
  • ad hoc to paper Clipping negative eigenvalues of the predicted matrix to zero preserves the information needed for portfolio optimization.
    Section 3.5 projects the predicted matrix to PSD by zeroing negative eigenvalues; the paper does not test whether this projection distorts the downside-risk signal that the semi-covariance matrix is supposed to capture.
  • domain assumption The one-month test window (February 12 to March 12, 2024) is representative of volatile market conditions and sufficient to compare models.
    Section 4.1 defines the test set as a single month; the paper generalizes from this window to claims about 'volatile market conditions' without any evidence that this month was volatile or representative.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dynamic ETF Portfolio Optimization Using enhanced Transformer-Based Models for Covariance and Semi-Covariance Prediction(Work in Progress)." pith.science (2026). https://pith.science/paper/C536TU5P

@misc{pith2026241119649,
  author       = {Pith},
  title        = {Pith review of: Dynamic ETF Portfolio Optimization Using enhanced Transformer-Based Models for Covariance and Semi-Covariance Prediction(Work in Progress)},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/C536TU5P}},
  note         = {Machine review of arXiv:2411.19649}
}
read the original abstract

This study explores the use of Transformer-based models to predict both covariance and semi-covariance matrices for ETF portfolio optimization. Traditional portfolio optimization techniques often rely on static covariance estimates or impose strict model assumptions, which may fail to capture the dynamic and non-linear nature of market fluctuations. Our approach leverages the power of Transformer models to generate adaptive, real-time predictions of asset covariances, with a focus on the semi-covariance matrix to account for downside risk. The semi-covariance matrix emphasizes negative correlations between assets, offering a more nuanced approach to risk management compared to traditional methods that treat all volatility equally. Through a series of experiments, we demonstrate that Transformer-based predictions of both covariance and semi-covariance significantly enhance portfolio performance. Our results show that portfolios optimized using the semi-covariance matrix outperform those optimized with the standard covariance matrix, particularly in volatile market conditions. Moreover, the use of the Sortino ratio, a risk-adjusted performance metric that focuses on downside risk, further validates the effectiveness of our approach in managing risk while maximizing returns. These findings have important implications for asset managers and investors, offering a dynamic, data-driven framework for portfolio construction that adapts more effectively to shifting market conditions. By integrating Transformer-based models with the semi-covariance matrix for improved risk management, this research contributes to the growing field of machine learning in finance and provides valuable insights for optimizing ETF portfolios.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

47 extracted references · 28 canonical work pages

  1. [1]

    , Alzoubi , O

    bchapter Alissa , K. , Alzoubi , O. : Financial sentiment analysis based on transformers and majority voting . In: 19th IEEE/ACS International Conference on Computer Systems and Applications, AICCSA 2022, Abu Dhabi, United Arab Emirates, December 5-8, 2022 , pp. 1 -- 4 . IEEE , ??? ( 2022 ). 10.1109/AICCSA56895.2022.10017941 . https://doi.org/10.1109/AICC...

    work page arXiv 2022

  2. [2]

    , : Estimating covariance matrices in high-dimensional finance

    barticle Balakrishnan , S. , : Estimating covariance matrices in high-dimensional finance . Journal of Financial Econometrics 18 ( 3 ), 593 -- 623 ( 2020 ) 10.1093/jjfinec/nby010 barticle

    work page doi:10.1093/jjfinec/nby010 2020

  3. [3]

    , Fehr , S

    barticle Bentz , R. , Fehr , S. : Positive semi-definiteness of covariance matrices in financial optimization . Mathematics of Finance 25 ( 3 ), 524 -- 536 ( 2009 ) 10.1111/j.1468-0327.2009.00315.x barticle

    work page arXiv 2009

  4. [4]

    , Litterman , R

    bbook Black , F. , Litterman , R. : Global Portfolio Optimization vol. 48 , pp. 28 -- 43 . Financial Analysts Journal , ??? ( 1992 ). 10.2469/faj.v48.n5.28 bbook

    doi:10.2469/faj.v48.n5.28 1992

  5. [5]

    , Levina , E

    barticle Bickel , P.J. , Levina , E. : Regularized estimation of large covariance matrices . The Annals of Statistics 36 ( 1 ), 199 -- 227 ( 2008 ) 10.1214/009053607000000945 barticle

    work page doi:10.1214/009053607000000945 2008

  6. [6]

    : Generalized autoregressive conditional heteroskedasticity

    barticle Bollerslev , T. : Generalized autoregressive conditional heteroskedasticity . Journal of Econometrics 31 ( 3 ), 307 -- 327 ( 1986 ) 10.1016/0304-4076(86)90063-1 barticle

    doi:10.1016/0304-4076(86)90063-1 1986

  7. [7]

    : A conditionally heteroskedastic time series model for speculative prices and rates of return

    barticle Bollerslev , T. : A conditionally heteroskedastic time series model for speculative prices and rates of return . Review of Economics and Statistics 71 ( 3 ), 542 -- 547 ( 1988 ) 10.2307/1927997 barticle

    work page doi:10.2307/1927997 1988

  8. [8]

    : Modelling the coherence of returns in financial markets

    barticle Bollerslev , T. : Modelling the coherence of returns in financial markets . Journal of Finance 45 ( 2 ), 307 -- 327 ( 1990 ) 10.1111/j.1540-6261.1990.tb02410.x barticle

    arXiv 1990

Show all 47 references
  1. [9]

    : Forecasting volatility and correlations in financial markets with machine learning techniques

    barticle Bollerslev , T. : Forecasting volatility and correlations in financial markets with machine learning techniques . Journal of Econometrics 209 ( 1 ), 64 -- 81 ( 2018 ) 10.1016/j.jeconom.2018.03.003 barticle

    2018 doi

  2. [10]

    , Silverstein , J.W

    bbook Bai , Z. , Silverstein , J.W. : Spectral Analysis of Large Dimensional Random Matrices . Springer , ??? ( 2010 ) bbook

    2010

  3. [11]

    , Vandenberghe , L

    bbook Boyd , S. , Vandenberghe , L. : Convex Optimization . Cambridge University Press , ??? ( 2004 ) bbook

    2004

  4. [12]

    , bingjia , X

    botherref chu , L. , bingjia , X. , qiping , Y. : UnetTSF: A Better Performance Linear Complexity Time Series Prediction Model (2024) botherref

    2024

  5. [13]

    , Kozen , D

    botherref Dahlqvist , F. , Kozen , D. : Semantics of higher-order probabilistic programs with conditioning. CoRR abs/1902.11189 (2019) https://arxiv.org/abs/1902.11189 1902.11189 botherref

    2019 arXiv

  6. [14]

    : Multivariate garch models

    barticle Engle , R.F. : Multivariate garch models . Journal of Econometrics 39 ( 4 ), 347 -- 357 ( 1995 ) 10.1016/0304-4076(95)01711-8 barticle

    1995 doi

  7. [15]

    : Dynamic conditional correlation: A simple class of multivariate garch models

    barticle Engle , R.F. : Dynamic conditional correlation: A simple class of multivariate garch models . Journal of Business & Economic Statistics 20 ( 3 ), 339 -- 350 ( 2002 ) barticle

    2002

  8. [16]

    : Risk optimization in portfolio management: An application of semi-covariance

    barticle Estrada , J. : Risk optimization in portfolio management: An application of semi-covariance . Journal of Portfolio Management 26 ( 4 ), 14 -- 26 ( 2000 ) 10.3905/jpm.2000.319852 barticle

  9. [17]

    : The semi-covariance matrix: A more intuitive alternative to the covariance matrix in portfolio optimization

    barticle Estrada , J. : The semi-covariance matrix: A more intuitive alternative to the covariance matrix in portfolio optimization . Journal of Portfolio Management 28 ( 1 ), 13 -- 22 ( 2002 ) 10.3905/jpm.2002.319491 barticle

    2002

  10. [18]

    : Efficient Capital Markets: A Review of Theory and Empirical Work vol

    bbook Fama , E.F. : Efficient Capital Markets: A Review of Theory and Empirical Work vol. 25 , pp. 383 -- 417 ( 1970 ). 10.2307/2325486 bbook

    1970 doi

  11. [19]

    , French , K.R

    barticle Fama , E.F. , French , K.R. : The cross-section of expected stock returns . The Journal of Finance 47 ( 2 ), 427 -- 465 ( 1992 ) 10.1111/j.1540-6261.1992.tb04398.x barticle

    1992

  12. [20]

    , French , K.R

    barticle Fama , E.F. , French , K.R. : Common risk factors in the returns on stocks and bonds . Journal of Financial Economics 33 ( 1 ), 3 -- 56 ( 1993 ) 10.1016/0304-405X(93)90023-5 barticle

    1993 doi

  13. [21]

    , French , K.R

    barticle Fama , E.F. , French , K.R. , Booth , D.G. , Sinquefield , R. : Differences in the risks and returns of nyse and nasd stocks . Financial Analysts Journal 49 ( 1 ), 37 -- 41 ( 1993 ). Accessed 2024-04-02 barticle

    1993

  14. [22]

    , Hastie , T

    barticle Friedman , J. , Hastie , T. , Tibshirani , R. : Sparse inverse covariance estimation with the graphical lasso . Biostatistics 9 ( 3 ), 432 -- 441 ( 2008 ) 10.1093/biostatistics/kxm045 barticle

    2008 doi

  15. [23]

    , Bengio , Y

    bbook Goodfellow , I. , Bengio , Y. , Courville , A. : Deep Learning . MIT Press , ??? ( 2016 ) bbook

    2016

  16. [24]

    , Kumar , A

    barticle Goetzmann , W.N. , Kumar , A. , Nuno , M. : The performance of sector-specific etfs: Evidence and implications for asset allocation . Financial Analysts Journal 73 ( 2 ), 65 -- 76 ( 2017 ) 10.2469/faj.v73.n2.4 barticle

    2017 doi

  17. [25]

    : Computing the nearest correlation matrix—a problem from finance

    barticle Higham , N.J. : Computing the nearest correlation matrix—a problem from finance . IMA Journal of Numerical Analysis 8 ( 3 ), 231 -- 243 ( 1988 ) 10.1093/imanum/8.3.231 barticle

    1988 doi

  18. [26]

    , Izumi , K

    barticle Hirano , M. , Izumi , K. , Sakaji , H. : STBM+: advanced stochastic trading behavior model for financial markets using residual blocks or transformers . New Gener. Comput. 40 ( 1 ), 7 -- 24 ( 2022 ) 10.1007/S00354-021-00145-Z barticle

    2022 doi

  19. [27]

    , Polson , N.G

    barticle Heaton , J.B. , Polson , N.G. , Witte , T.L. : Deep learning in finance . Journal of Financial Data Science 1 , 1 -- 27 ( 2016 ) 10.3905/jfds.2016.1.1.001 barticle

    2016 doi

  20. [28]

    , Li , W

    barticle Jiang , G. , Li , W. : Short-selling and market regulation in china . China Finance Review International 5 ( 4 ), 356 -- 372 ( 2015 ) 10.1108/CFRI-07-2014-0094 barticle

    2015 doi

  21. [29]

    : Global portfolio optimization: The minimum variance approach

    barticle Jorion , P. : Global portfolio optimization: The minimum variance approach . Financial Analysts Journal 42 ( 1 ), 61 -- 67 ( 1986 ) 10.2469/faj.v42.n1.61 barticle

    1986 doi

  22. [30]

    , Kaiser , L

    bchapter Kitaev , N. , Kaiser , L. , Levskaya , A. : Reformer: The efficient transformer . In: 8th International Conference on Learning Representations, ICLR 2020, Addis Ababa, Ethiopia, April 26-30, 2020 . OpenReview.net , ??? ( 2020 ). https://openreview.net/forum?id=rkgNKkH...

    2020

  23. [31]

    , Wolf , M

    barticle Ledoit , O. , Wolf , M. : Honey, i shrunk the sample covariance matrix . The Journal of Portfolio Management 30 ( 4 ), 110 -- 119 ( 2004 ) 10.3905/jpm.2004.110 barticle

    2004 doi

  24. [32]

    : Portfolio selection

    barticle Markowitz , H. : Portfolio selection . The Journal of Finance 7 ( 1 ), 77 -- 91 ( 1952 ) 10.2307/2975974 barticle

    1952 doi

  25. [33]

    : The efficient portfolio: A brief history

    barticle Markowitz , H. : The efficient portfolio: A brief history . The Journal of Financial and Quantitative Analysis 44 ( 3 ), 671 -- 686 ( 2009 ) 10.1017/S0022109009990174 barticle

    2009 doi

  26. [34]

    : The expected return-covariance matrix in portfolio optimization

    barticle Michaud , R. : The expected return-covariance matrix in portfolio optimization . Journal of Financial and Quantitative Analysis 24 ( 2 ), 179 -- 187 ( 1989 ) 10.2307/2330950 barticle

    1989 doi

  27. [35]

    , Zhang , T

    barticle Ma , S. , Zhang , T. , Zhao , Y. , Kang , Y. , Bai , P. : TCLN: A transformer-based conv-lstm network for multivariate time series forecasting . Appl. Intell. 53 ( 23 ), 28401 -- 28417 ( 2023 ) 10.1007/S10489-023-04980-Z barticle

    2023 doi

  28. [36]

    : Introductory Lectures on Convex Optimization: A Basic Course

    bbook Nesterov , Y. : Introductory Lectures on Convex Optimization: A Basic Course . Kluwer Academic Publishers , ??? ( 1994 ) bbook

    1994

  29. [37]

    , West , K.D

    barticle Newey , W.K. , West , K.D. : A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix . Econometrica 55 ( 3 ), 703 -- 708 ( 1987 ). Accessed 2024-11-06 barticle

    1987

  30. [38]

    : The arbitrage theory of capital asset pricing

    barticle Ross , S.A. : The arbitrage theory of capital asset pricing . Journal of Economic Theory 13 ( 3 ), 341 -- 360 ( 1976 ) 10.1016/0022-0531(76)90046-6 barticle

    1976 doi

  31. [39]

    : Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead

    barticle Rudin , C. : Stop explaining black box machine learning models for high stakes decisions and use interpretable models instead . Nature Machine Intelligence 1 ( 5 ), 206 -- 215 ( 2019 ) 10.1038/s41586-019-0101-1 barticle

    2019 doi

  32. [40]

    : Capital asset prices: A theory of market equilibrium under conditions of risk

    barticle Sharpe , W.F. : Capital asset prices: A theory of market equilibrium under conditions of risk . The Journal of Finance 19 ( 3 ), 425 -- 442 ( 1964 ) 10.2307/2977928 barticle

    1964 doi

  33. [41]

    , Shazeer , N

    botherref Vaswani , A. , Shazeer , N. , Parmar , N. , Uszkoreit , J. , Jones , L. , Gomez , A.N. , Kaiser , L. , Polosukhin , I. : Attention is all you need. CoRR abs/1706.03762 (2017) https://arxiv.org/abs/1706.03762 1706.03762 botherref

    2017 arXiv

  34. [42]

    , Shazeer , N

    botherref Vaswani , A. , Shazeer , N. , Parmar , N. , Uszkoreit , J. , Jones , L.K. , Nips , A. , Polosukhin , I. : Attention is all you need. Proceedings of the 31st International Conference on Neural Information Processing Systems, 6000--6010 (2017) 10.5555/3295222.3295349 botherref

    2017

  35. [43]

    , Chen , Y

    barticle Wang , C. , Chen , Y. , Zhang , S. , Zhang , Q. : Stock market index prediction using deep transformer model . Expert Syst. Appl. 208 , 118128 ( 2022 ) 10.1016/J.ESWA.2022.118128 barticle

    2022

  36. [44]

    , Xu , J

    botherref Wu , H. , Xu , J. , Wang , J. , Long , M. : Autoformer: Decomposition transformers with auto-correlation for long-term series forecasting. CoRR abs/2106.13008 (2021) https://arxiv.org/abs/2106.13008 2106.13008 botherref

    2021 arXiv

  37. [45]

    , Ma , Z

    botherref Zhou , T. , Ma , Z. , Wen , Q. , Wang , X. , Sun , L. , Jin , R. : Fedformer: Frequency enhanced decomposed transformer for long-term series forecasting. CoRR abs/2201.12740 (2022) https://arxiv.org/abs/2201.12740 2201.12740 botherref

    2022 arXiv

  38. [46]

    , Zhang , S

    botherref Zhou , H. , Zhang , S. , Peng , J. , Zhang , S. , Li , J. , Xiong , H. , Zhang , W. : Informer: Beyond efficient transformer for long sequence time-series forecasting. CoRR abs/2012.07436 (2020) https://arxiv.org/abs/2012.07436 2012.07436 botherref

    2020 arXiv

  39. [47]

    write newline

    " write newline "" before.all 'output.state := FUNCTION string.to.integer 't := t text.length 'k := #1 'char.num := t char.num #1 substring 's := s is.num s "." = or char.num k = not and char.num #1 + 'char.num := while char.num #1 - 'char.num := t #1 char.num substring FUNCTI...

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.