Pith. sign in

REVIEW 3 major objections 5 minor 17 cited by

Qlib: An AI-oriented Quantitative Investment Platform

T0 review · 3 major / 5 minor · reviewed 2026-08-27 · deepseek-v4-flash

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

Pith's one-line read Qlib is an open-source platform for the full AI-driven quantitative workflow, with a flat-file engine that turns a typical factor-construction task into seconds.

desk verdict Qlib is a real open-source AI-quant platform with a sensible modular design; the paper's headline speedup is a warm-cache number, so the performance claim needs repair, not the whole paper. read the letter →

arxiv 2009.11189 v1 pith:LYXZF7HB submitted 2020-09-22 q-fin.GN cs.LGq-fin.PM

Xiao Yang , Weiqing Liu , Dong Zhou , Jiang Bian , Tie-Yan Liu This is my paper · ORCID
classification q-fin.GNcs.LGq-fin.PM
keywords quantitativeinvestmentmachinelearningdatainfrastructureflat-filedatabaseexpressionenginecachesystemportfoliomanagementbacktesting
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

Qlib is an open-source quantitative investment platform designed around the workflow of a modern AI-driven researcher: it connects data serving, feature construction, model training and management, portfolio generation, order execution, and performance analysis into one modular pipeline. The paper argues that existing tools, whether general-purpose databases, time-series databases, or array-file formats, either lack this workflow coverage or pay heavy overhead in format conversion, and that Qlib's dedicated storage closes that gap. Its central performance claim is that a typical factor-construction task over 800 daily-changing stocks finishes in about 7 seconds with Qlib's caches enabled, versus roughly 3 to 6 minutes for HDF5, MySQL, MongoDB, and InfluxDB. The broader claim is that this infrastructure makes AI research topics in finance feasible rather than merely slower.

What carries the argument

The load-bearing object is the flat-file database: a tree of fixed-width binary files, one per instrument and attribute, with a shared calendar file and a 4-byte start-time index in each data file. Because values are laid out contiguously in time order, reads can slice directly into array-based scientific formats without format transformation, and new data can be appended efficiently. On top of this sit the expression engine, which parses factor formulas such as Bollinger bands into syntax trees and caches subexpression results, and a two-level disk cache that stores already computed factors and already combined arrays so repeated research queries skip both computation and concatenation. Together these mechanisms convert a task that would otherwise spend most of its time loading, evaluating expressions, and combining arrays into a near-memory-speed read when caches are warm.

What would settle it

Run the same 14-factor, 800-stock task from a cold start, including the time to build all expression and dataset caches, on a stated hardware configuration; if Qlib with caches enabled is not faster overall than the array-file baseline, the central performance claim collapses. The 'first open-source full-workflow platform' claim would also be refuted by any earlier open-source platform with comparable coverage.

Watch

Extended reading notes

Core claim

The paper claims Qlib is the first open-source platform that accommodates the complete workflow of a modern quantitative researcher in the age of AI, and that its high-performance infrastructure outperforms most existing solutions on a typical data-processing task. On the workflow side, Qlib's modular design treats each stage, including the Data Server, Data Enhancement, Model Creator, Model Manager, Model Ensemble, Portfolio Generator, Order Executor, and Analysers, as an interface with default implementations, so researchers can replace one stage without rebuilding the rest, and it supports dynamic updates and interactive paradigms such as reinforcement learning. On the infrastructure side, the paper reports that a task of building 14 derived features from daily OHLCV data over 800 stocks from 2007 to 2020 takes about 7.4 seconds with Qlib's expression and dataset caches enabled, compared with 184 to 368 seconds for HDF5, MySQL, MongoDB, and InfluxDB, and it attributes the gain to compact fixed-width binary storage, byte-indexed time alignment, and a two-level cache that avoids recomputing shared expressions and recombining arrays.

Load-bearing premise

The performance conclusions assume the benchmark task is representative of real quantitative research and that warm-cache operation is the appropriate mode, but the paper does not report the cost of building those caches, the hardware used, or the exact data source.

Editorial extensions

If this is right

  • A quantitative researcher can prototype an end-to-end AI strategy, from raw data to executed orders and analysis reports, from a config file rather than assembling and maintaining separate tools.
  • The expression engine makes factor search practical: new features can be written as strings and cached subexpressions reused, lowering the cost of automated factor discovery.
  • The responsive order-execution simulator, rather than a one-shot backtester, enables reinforcement learning and other interactive paradigms to train against market feedback.
  • Dynamic workflows become supported on the same platform: models and strategies can be retrained or retuned on rolling data, with hyperparameter search biased toward the previous best setting.
  • For teams working with large datasets, the flat-file design removes database format conversion overhead, making high-frequency research data practical on a single machine.

Reading between the lines

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

  • If warm-cache performance reflects repeated research loops, the biggest practical win comes from cache reuse: teams that re-run the same feature set over rolling windows may see larger gains than one-shot benchmark users, which the paper states as a design goal but does not measure directly.
  • The same flat-file storage recipe transfers to any append-only, time-indexed scientific dataset, such as sensor logs or telemetry, where array-oriented analytics dominate; the paper only demonstrates finance data.
  • A fairer cross-platform comparison would report cold-cache construction cost and hardware; without that, the headline seconds describe a repeat user's typical run, not a new user's first run.
  • The 'first open-source platform' claim is only as strong as the definition of workflow scope; an explicit coverage matrix of existing open-source tools would make the claim checkable.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Request a human review

A listed scientist reviews the paper for a fee and the review publishes here regardless of verdict. See the reviewers or get listed.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. Qlib: An AI-oriented Quantitative Investment Platform describes an open-source Python framework for quantitative investment built around modern machine-learning workflows. The paper motivates the platform by the shift from traditional factor/backtest pipelines to AI-driven, data-intensive research, and introduces a modular architecture (data server, data enhancement, model creation/ensemble/management, portfolio generation, order execution, analysis) plus a flat-file binary storage system, an expression engine for factor computation, a two-level disk cache, and hyperparameter tuning. The evaluation constructs a 14-factor dataset from daily OHLCV data for 800 stocks over 2007-2020 and compares Qlib (with combinations of expression and dataset caches) against HDF5, MySQL, MongoDB, and InfluxDB; Qlib with both caches enabled reports 7.4 seconds total versus 184.4 seconds for HDF5. The paper concludes that Qlib's infrastructure outperforms most existing solutions on typical quant data-processing tasks.

Significance. If substantiated, the platform is a useful open-source contribution: it offers a complete modular workflow, an array-oriented storage design, and engineering mechanisms (LRU memory cache, two-level disk cache, expression engine) that directly address the needs of factor-based research with AI models. The code is publicly available, which is a concrete strength and facilitates verification. The performance story, however, is the principal technical contribution, and as reported it is not yet convincing: the fastest configuration exploits precomputed caches whose construction cost is not measured, the benchmark setup is not sufficiently specified for reproduction, and the claimed representativeness of the workload is not established. The paper currently reads more as a system description with a preliminary benchmark than as a validated performance study.

major comments (3)
  1. [Section 4.2, Table 1] The headline result for Qlib +E +D is a warm-cache measurement: the text states that this configuration 'just reads the existing cache and almost computes nothing' (§4.2). The time needed to build the expression cache and the dataset cache is not reported anywhere in the paper. Because the HDF5, MySQL, MongoDB, and InfluxDB baselines are timed over the full pipeline (including expression computation and data combination), the approximately 25x speedup versus HDF5 conflates cache reuse with the storage/query design itself. The paper should report cold-cache, end-to-end timings (including cache construction) or clearly separate one-time cache-building cost from repeated-query cost, and should restrict the 'outperforms most existing solutions' claim to the mode that is actually measured.
  2. [Section 4.2, Table 1] The benchmark is not reproducible as reported. The paper does not state the hardware configuration (CPU model, memory, disk type, OS), software versions, the exact data source, or the instrument universe; it only says that the data are 'basic OHLCV daily data of a stock market' with a daily pool of 800 stocks. In addition, no information is given about the number of runs behind the reported means and standard deviations. These details are necessary for interpreting the error bars and for independently verifying the comparison, especially since the performance evaluation is the load-bearing evidence for the paper's main claim.
  3. [Section 4.2] The evaluation's workload is asserted to be 'a typical task in quantitative investment', but the paper gives no evidence for this. The benchmark uses only 14 fixed factors, daily-frequency data, and a single dataset-construction pass; it does not exercise the data update/append path, ad-hoc factor exploration (where expressions change frequently), or repeated queries with incremental data, all of which Section 3.3 identifies as important in real quant research. The authors should either provide evidence that the chosen mix of load, expression computation, filtering, and combination is representative or soften the scope of the performance conclusion to repeated identical queries on a static dataset.
minor comments (5)
  1. [Section 4.1] The reference 'List ??' appears to be an unresolved placeholder; the configuration example is shown in Figure 4, so the text should point to the figure instead.
  2. [Section 4.2] The table layout is hard to read: several cells are blank or misaligned, and it is not clear which pipeline steps are applicable to each system; for Qlib +E +D, the 64-CPU entry should be marked explicitly as unchanged (or 'not applicable') rather than left blank.
  3. [Sections 3.2 and 4.2] There are several typos: 'Model Creater' (§3.2), 'As we can seen' and 'Tabel 1' (§4.2), and 'opensource' (§4.3).
  4. [Section 3.4] The hyperparameter tuning distribution is introduced with an inline formula; adding an equation number and a one-line definition of \phi_{\theta_{\mathrm{prev}},\sigma^2} would improve clarity.
  5. [Section 2.2] The claim that Qlib is 'the first open-source platform' would benefit from a broader related-work survey or a more qualified wording; the current short list of tools does not by itself establish uniqueness.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: this is a systems benchmark paper with no fitted prediction; the warm-cache benchmark concern is an empirical-validity gap, not circularity.

full rationale

Qlib is a systems paper; its central claims are that it is the first open-source AI-oriented quant platform and that its infrastructure outperforms existing storage solutions on a typical dataset-construction task. Neither claim is derived from an equation or fitted parameter. Table 1 reports measured wall-clock times, and the fastest Qlib configuration is explicitly disclosed as reading an existing cache: the paper states that Qlib +E +D 'just reads the existing cache and almost computes nothing' (§4.2). The omission of cache-construction cost is a benchmark external-validity concern, not a circular reduction, because the reported number is an honest measurement of a narrow configuration. The self-citations (Wang et al. 2019a; Yang et al. 2019) are used only to motivate dynamic model and strategy modules and do not constitute evidence for the platform or performance claims. No uniqueness theorem, ansatz, or fitted parameter is imported from prior work. Therefore no circular step can be exhibited, and the appropriate finding is no significant circularity.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The paper does not fit data or introduce theoretical constructs; it rests on domain assumptions about typical quantitative work flows and fair benchmarking.

assumptions (4)
  • domain assumption Financial data has extremely low signal-to-noise ratio, making domain-adapted machine learning necessary.
    Used in Section 2.1 to motivate the platform's guidance features; not empirically demonstrated in the paper.
  • domain assumption The benchmark workload (14 OHLCV-derived factors, 800 stocks, 2007-2020) is representative of typical quantitative research data processing.
    Section 4.2 basis for performance claims; no sensitivity analysis to workload variations is provided.
  • domain assumption The compared storage solutions were configured fairly and optimally by the authors.
    Section 4.2; exact versions and configurations of HDF5, MySQL, MongoDB, and InfluxDB are not specified.
  • domain assumption Warm-cache operation is the relevant mode for real users.
    The Qlib +E +D result in Table 1 dominates the comparison, but the cost of building the caches is not reported, so the comparison may not reflect typical first-run usage.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Qlib: An AI-oriented Quantitative Investment Platform." pith.science (2026). https://pith.science/paper/LYXZF7HB

@misc{pith2026200911189,
  author       = {Pith},
  title        = {Pith review of: Qlib: An AI-oriented Quantitative Investment Platform},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LYXZF7HB}},
  note         = {Machine review of arXiv:2009.11189}
}
read the original abstract

Quantitative investment aims to maximize the return and minimize the risk in a sequential trading period over a set of financial instruments. Recently, inspired by rapid development and great potential of AI technologies in generating remarkable innovation in quantitative investment, there has been increasing adoption of AI-driven workflow for quantitative research and practical investment. In the meantime of enriching the quantitative investment methodology, AI technologies have raised new challenges to the quantitative investment system. Particularly, the new learning paradigms for quantitative investment call for an infrastructure upgrade to accommodate the renovated workflow; moreover, the data-driven nature of AI technologies indeed indicates a requirement of the infrastructure with more powerful performance; additionally, there exist some unique challenges for applying AI technologies to solve different tasks in the financial scenarios. To address these challenges and bridge the gap between AI technologies and quantitative investment, we design and develop Qlib that aims to realize the potential, empower the research, and create the value of AI technologies in quantitative investment.

Figures

Figures reproduced from arXiv: 2009.11189 by the authors.

Figure 1
Figure 1. modules and a typical workflow built with Qlib [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The description of the flat-file database; the left part is the [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 4
Figure 4. A Configuration example of CDPE The task for the solutions is to create a dataset from the basic OHLCV7 daily data of a stock market, which involves data query and processing. The final dataset consists of 14 factors/features derived from OHLCV data(e.g. ”Std($close, 5)/$close”). The time of the data ranges from 1/1/2007 to 1/1/2020. The stock pool consists of 800 stocks each day, which changes daily. Besides the co… view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 17 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Centering Drives Normalization Gains: Price-Offset Nuisances in Cross-Sectional Return Prediction

    cs.CE 2026-09 accept novelty 8.0 of 10

    Centering price channels, not amplitude scaling or encoder choice, drives instance normalization gains in cross-sectional return prediction.

  2. Cognitive Alpha Mining via LLM-Driven Code-Based Evolution

    cs.CL 2025-11 unverdicted novelty 7.0 of 10

    CogAlpha combines LLM reasoning with code-level evolutionary search to discover financial alphas that show higher predictive accuracy and generalization than prior methods on five stock datasets.

  3. GoAnt: Quality-Diversity Multi-Agent Search for Alpha Factor Discovery in Market Microstructure Data

    cs.AI 2026-09 conditional novelty 6.0 of 10

    An adaptive-capacity quality-diversity archive with decoupled worker roles and a distilled routing Queen finds more high-quality, mutually non-redundant trading factors than static-topology baselines under matched eva...

  4. A Compact Selective State-Space Model for Cross-Sectional Stock Return Ranking from Raw Intraday Bars

    cs.CE 2026-08 conditional novelty 6.0 of 10

    STRATA, a compact selective state-space model on raw intraday bars, ranks next-day stock returns better than six parameter-matched sequence baselines after style residualisation, but its executable-session spread is s...

  5. AlphaG-OPD: Reliability-Gated Sibling Counterfactuals for On-Policy Distillation in Symbolic Alpha Factor Discovery

    cs.LG 2026-08 conditional novelty 6.0 of 10

    A reliability-gated sibling-counterfactual distillation objective improves GFlowNet-based symbolic alpha factor discovery across four equity markets under a fixed evaluation budget.

  6. AlphaSchema: Exploring the Space of Trading Semantics for LLM-Based Alpha Mining

    cs.AI 2026-07 conditional novelty 6.0 of 10

    Searching over explicit trading-semantics plans (event/context/qualities/direction/output), instead of factor code, yields competitive alpha pools and makes the search space controllable and learnable.

  7. AlphaForgeBench: Benchmarking End-to-End Trading Strategy Design with Large Language Models

    q-fin.TR 2026-02 conditional novelty 6.0 of 10

    LLMs are unreliable when asked to emit buy/sell/hold actions, so this paper benchmarks them as code-writing quantitative researchers whose generated strategies are backtested deterministically.

  8. AlphaEval: A Comprehensive and Efficient Evaluation Framework for Formula Alpha Mining

    cs.AI 2025-08 conditional novelty 6.0 of 10

    AlphaEval scores alpha mining models on prediction, stability, robustness, logic, and diversity, replacing backtests with fast parallel metrics that the paper claims align with backtest outcomes.

  9. Learning from Expert Factors: Trajectory-level Reward Shaping for Formulaic Alpha Mining

    cs.LG 2025-07 conditional novelty 6.0 of 10

    A trajectory-level reward shaping method for RL-based formulaic alpha mining uses exact subsequence matching against expert formulas and reward centering to accelerate training and slightly improve mined factors.

  10. Learning Universal Multi-level Market Irrationality Factors to Improve Stock Return Forecasting

    cs.LG 2025-02 conditional novelty 6.0 of 10

    UMI adds learned stock-level mispricing residuals and market-level synchrony representations to stock return forecasting models, improving IC and Sharpe ratio on US and China markets.

  11. BPQP: A Differentiable Convex Optimization Framework for Efficient End-to-End Learning

    cs.LG 2024-11 conditional novelty 6.0 of 10

    BPQP reformulates the backward pass of differentiable convex optimization layers as an equality-constrained quadratic program, allowing fast ADMM-based solvers to compute gradients.

  12. Forecast Collapse in Time-Series Foundation Models

    cs.LG 2026-08 conditional novelty 5.0 of 10

    Forecast collapse in time-series models comes from low target predictability and per-series objectives that ignore cross-series structure; a composite MSE-correlation loss, CalibRank, nearly triples cross-sectional co...

  13. F$^2$Agent: Financial Fusion of Agentic Intelligence for Multimodal Trading

    cs.MA 2026-08 reject novelty 5.0 of 10

    F2Agent, a hierarchy of specialized LLM and Transformer agents with adaptive cross-modal attention and consistency regularization, is reported to beat 16 trading baselines on six assets, though appendix results from a...

  14. Evolutionary Factor Searching for Sparse Portfolio Optimization Using Large Language Models

    q-fin.PM 2025-07 reject novelty 5.0 of 10

    An LLM-driven evolutionary search that generates and refines alpha factors outperforms statistical and optimization baselines on sparse, small-size portfolios in backtests.

  15. QuantBench: Benchmarking AI Methods for Quantitative Investment

    q-fin.CP 2025-04 conditional novelty 5.0 of 10

    QuantBench introduces a unified, industry-aligned benchmark platform for evaluating AI methods across the full quantitative investment pipeline, with data, models, and empirical comparisons.

  16. Towards Autonomous Formulaic Alpha Discovery: An Evolutionary Computation Perspective

    cs.NE 2026-08 conditional novelty 4.0 of 10

    A review that reframes automated trading-signal (alpha) discovery as noisy, dynamic, multiobjective evolutionary optimization, and proposes six-component and eight-dimension frameworks for comparing and evaluating methods.

  17. GRU-PFG: Extract Inter-Stock Correlation from Stock Factors with Graph Neural Network

    q-fin.CP 2024-11 conditional novelty 4.0 of 10

    GRU-PFG, a factor-only model, reports IC 0.134 on CSI300, surpassing the multi-source HIST model's 0.131.

Reference graph

Works this paper leans on

26 extracted references · 26 canonical work pages · cited by 17 Pith papers

  1. [1]

    Stock market volatility and learning,

    [Adam et al., 2016] Klaus Adam, Albert Marcet, and Juan Pablo Nicolini. Stock market volatility and learning,

    work page 2016

  2. [6]

    Enhancing stock movement prediction with adversarial training

    [Feng et al., 2019] Fuli Feng, Huimin Chen, Xiangnan He, Ji Ding, Maosong Sun, and Tat-Seng Chua. Enhancing stock movement prediction with adversarial training. In Proceedings of the 28th International Joint Conference on Artificial Intelligence, pages 5843–5849. AAAI Press,

    work page 2019

  3. [9]

    Olps: a toolbox for on-line portfolio selection.The Journal of Machine Learning Research, 17(1):1242–1246,

    [Li et al., 2016] Bin Li, Doyen Sahoo, and Steven CH Hoi. Olps: a toolbox for on-line portfolio selection.The Journal of Machine Learning Research, 17(1):1242–1246,

    work page 2016

  4. [10]

    pandas: a foundational python library for data analysis and statistics

    [McKinney, 2011] Wes McKinney. pandas: a foundational python library for data analysis and statistics. Python for High Performance and Scientific Computing, 14,

    work page 2011

  5. [12]

    [MySQL, 2001] AB MySQL. Mysql,

    work page 2001

  6. [17]

    Generating trading rules on the stock mar- kets with genetic programming

    [Potvin et al., 2004] Jean-Yves Potvin, Patrick Soriano, and Maxime Vall´ee. Generating trading rules on the stock mar- kets with genetic programming. Computers & Operations Research, 31(7):1033–1047,

    work page 2004

  7. [18]

    Quantitative equity portfolio manage- ment: modern techniques and applications

    [Qian et al., 2007] Edward E Qian, Ronald H Hua, and Eric H Sorensen. Quantitative equity portfolio manage- ment: modern techniques and applications . CRC Press,

    work page 2007

  8. [21]

    Barra’s risk models

    [Sheikh, 1996] Aamir Sheikh. Barra’s risk models. Barra Research Insights, pages 1–24,

    work page 1996

Show all 26 references
  1. [23]

    Conservative or aggressive? confidence- aware dynamic portfolio construction

    [Wang et al., 2019a] Lewen Wang, Weiqing Liu, Xiao Yang, and Jiang Bian. Conservative or aggressive? confidence- aware dynamic portfolio construction. In 2019 IEEE Global Conference on Signal and Information Processing (GlobalSIP), pages 1–5. IEEE,

    2019

  2. [24]

    Stock market index prediction using deep neural network ensemble

    [Yang et al., 2017] Bing Yang, Zi-Jia Gong, and Wenqi Yang. Stock market index prediction using deep neural network ensemble. In 2017 36th Chinese Control Confer- ence (CCC), pages 3882–3887. IEEE,

    2017

  3. [25]

    A divide-and-conquer frame- work for attention-based combination of multiple invest- ment strategies

    [Yang et al., 2019] Xiao Yang, Weiqing Liu, Lewen Wang, Cheng Qu, and Jiang Bian. A divide-and-conquer frame- work for attention-based combination of multiple invest- ment strategies. In 2019 IEEE Global Conference on Sig- nal and Information Processing (GlobalSIP) , pages 1–5. IEEE,

    2019

  4. [26]

    A deep learning ensemble approach for crude oil price fore- casting

    [Zhao et al., 2017] Yang Zhao, Jianping Li, and Lean Yu. A deep learning ensemble approach for crude oil price fore- casting. Energy Economics, 66:9–16, 2017

    2017

  5. [1996]

    A perspective view and survey of meta-learning

    [Vilalta and Drissi, 2002] Ricardo Vilalta and Youssef Drissi. A perspective view and survey of meta-learning. Artificial intelligence review, 18(2):77–95,

    2002

  6. [1997]

    A guide to NumPy, vol- ume

    [Oliphant, 2006] Travis E Oliphant. A guide to NumPy, vol- ume

    2006

  7. [1999]

    Bollinger on Bollinger bands

    [Bollinger, 2002] John Bollinger. Bollinger on Bollinger bands. McGraw Hill Professional,

    2002

  8. [2001]

    Time series databases and in- fluxdb

    [Naqvi et al., 2017] Syeda Noor Zehra Naqvi, Sofia Yfanti- dou, and Esteban Zim ´anyi. Time series databases and in- fluxdb. Studienarbeit, Universit´e Libre de Bruxelles,

    2017

  9. [2002]

    MongoDB: the definitive guide: powerful and scalable data storage

    [Chodorow, 2013] Kristina Chodorow. MongoDB: the definitive guide: powerful and scalable data storage . ” O’Reilly Media, Inc.”,

    2013

  10. [2004]

    101 formulaic al- phas

    [Kakushadze, 2016] Zura Kakushadze. 101 formulaic al- phas. Wilmott, 2016(84):72–81,

    2016

  11. [2006]

    Do the fama–french factors proxy for innovations in predictive variables? The Journal of Finance, 61(2):581–612,

    [Petkova, 2006] Ralitsa Petkova. Do the fama–french factors proxy for innovations in predictive variables? The Journal of Finance, 61(2):581–612,

    2006

  12. [2007]

    Ensemble deep learning for regression and time series forecasting

    [Qiu et al., 2014] Xueheng Qiu, Le Zhang, Ye Ren, Pon- nuthurai N Suganthan, and Gehan Amaratunga. Ensemble deep learning for regression and time series forecasting. In 2014 IEEE symposium on computational intelligence in ensemble learning (CIEL), pages 1–6. IEEE,

    2014

  13. [2011]

    Technical analysis of the fi- nancial markets: A comprehensive guide to trading meth- ods and applications

    [Murphy, 1999] John J Murphy. Technical analysis of the fi- nancial markets: A comprehensive guide to trading meth- ods and applications. Penguin,

    1999

  14. [2013]

    Deep direct reinforce- ment learning for financial signal representation and trad- ing

    8https://github.com/microsoft/qlib/ [Deng et al., 2016] Yue Deng, Feng Bao, Youyong Kong, Zhiquan Ren, and Qionghai Dai. Deep direct reinforce- ment learning for financial signal representation and trad- ing. IEEE transactions on neural networks and learning systems, 28(3):653–664,

    2016

  15. [2014]

    Financial time se- ries forecasting with deep learning: A systematic literature review: 2005-2019

    [Sezer et al., 2019] Omer Berat Sezer, Mehmet Ugur Gudelek, and Ahmet Murat Ozbayoglu. Financial time se- ries forecasting with deep learning: A systematic literature review: 2005-2019. arXiv preprint arXiv:1911.13288 ,

    2019 arXiv

  16. [2016]

    Using genetic algorithms to find technical trad- ing rules

    [Allen and Karjalainen, 1999] Franklin Allen and Risto Kar- jalainen. Using genetic algorithms to find technical trad- ing rules. Journal of financial Economics, 51(2):245–271,

    1999

  17. [2017]

    Is technical analysis in the foreign exchange mar- ket profitable? a genetic programming approach

    [Neely et al., 1997] Christopher Neely, Paul Weller, and Rob Dittmar. Is technical analysis in the foreign exchange mar- ket profitable? a genetic programming approach. Jour- nal of financial and Quantitative Analysis, 32(4):405–426,

    1997

  18. [2019]

    Why use quantlib

    [Firth, 2004] N Firth. Why use quantlib. Paper available at: http://www. quantlib. co. uk/publications/quantlib. pdf,

    2004

Pith tools

Reviewed August 27, 2026 · model on record in the stance chip above.