Pith. sign in

REVIEW 5 major objections 4 minor 30 references

Hybrid Vector Auto Regression and Neural Network Model for Order Flow Imbalance Prediction in High Frequency Trading

T0 review · 5 major / 4 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 A hybrid VAR-neural network model predicts order flow imbalance more accurately than either model alone.

desk verdict A routine hybrid VAR-FNN applied to OFI, with a validation protocol that cannot support its central claim of consistent outperformance. read the letter →

arxiv 2411.08382 v1 pith:5LCWJKAI submitted 2024-11-13 q-fin.CP q-fin.STq-fin.TR

Abdul Rahman , Neelesh Upadhye This is my paper · ORCID
classification q-fin.CPq-fin.STq-fin.TR
keywords orderflowimbalancehigh-frequencytradingvectorautoregressionfeedforwardneuralnetworkhybridmodelintensitylimitbookcryptocurrencymarketdata
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that predicting order flow imbalance (OFI), the net buying versus selling pressure in a limit order book, is better done by a two-stage hybrid than by either component alone. The hybrid first fits a vector autoregression to buy and sell order counts, then trains a small feedforward network on the residuals to capture nonlinear structure, and combines the two forecasts into a final OFI prediction. On two real cryptocurrency order books and one synthetic dataset, the authors report lower mean squared and absolute errors, higher $R^2$, and higher accuracy on a buy/sell/hold trading-intensity signal compared with standalone VAR and FNN models. If the claim holds, the model gives high-frequency traders a directly usable signal of which side of the market is under pressure.

What carries the argument

The central object is the residual-learning decomposition $OFI_t = OFI_t^{\mathrm{VAR}} + f(\varepsilon_t)$, where $\varepsilon_t$ is the vector of VAR residuals from modeling buy and sell orders and $f$ is a feedforward neural network trained on those residuals. The VAR supplies the linear, interpretable baseline; the FNN supplies the nonlinear correction; the final OFI prediction is the sum of the two. The same pipeline generates a trading-intensity signal by thresholding OFI into BUY, SELL, or HOLD, and that signal is scored for accuracy and precision.

What would settle it

Run the same VAR-FNN pipeline after choosing all hyperparameters on a separate tuning set, then evaluate on untouched BTCUSD, ETCUSDT, and synthetic datasets; if the hybrid no longer beats the standalone FNN on error metrics and intensity accuracy, the central claim fails. A dataset whose VAR residuals are pure noise should likewise erase the hybrid's advantage over VAR alone.

Watch

Extended reading notes

Core claim

The paper's central claim is that a hybrid Vector Auto Regression-Feedforward Neural Network (VAR-FNN) model achieves superior forecasting accuracy for Order Flow Imbalance (OFI) in high-frequency trading, on both synthetic and real Binance data, and that the same model yields better trading-intensity signals than standalone VAR or FNN models. The authors define OFI as the difference over a window between buy and sell order counts divided by their sum, so values lie in $[-1,1]$. Their hybrid first fits a VAR with lag order 2 to buy and sell order series, computes residuals by subtracting the VAR forecasts from the actual series, trains a 32-16-2 feedforward network with ReLU activations on those residuals, and combines the VAR-based OFI with the FNN-predicted residual OFI to make the final forecast. On the reported validation sets, the hybrid attains $R^2$ values of 0.997, 0.983, and 0.999 for BTCUSD, ETCUSDT, and synthetic data, with trading-intensity accuracy of 98.18%, 96.41%, and 99.77%.

Load-bearing premise

The reported superiority of the hybrid model assumes that the three validation datasets were independent of model selection, but Section 4.2.1 says 120 parameter combinations were evaluated on exactly those datasets to pick the optimal configuration, so the validation metrics are in-sample for the hyperparameters.

Editorial extensions

If this is right

  • If the central claim is right, practitioners get a cheap two-stage recipe: fit the linear VAR first, then learn only the residual structure, which reduces what the network must model.
  • The reported results imply that liquid cryptocurrency order books carry enough autocorrelation in OFI that a lag-2 VAR captures the linear core, with exploitable nonlinear structure left in the residuals.
  • Trading-intensity accuracy above 96% on real data implies the model can generate actionable BUY/SELL/HOLD labels at one-second horizons, not just point forecasts of OFI.
  • The near-saturation on the synthetic dataset ($R^2=0.999$, accuracy 99.77%) suggests the residual-learning mechanism recovers the data-generating process when nonlinearity is present.
  • The hybrid can serve as a natural baseline for future OFI forecasting work that tries to beat a linear-plus-nonlinear residual split.

Reading between the lines

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

  • If the result generalizes to other assets, the same residual-learning split could be applied to other noisy microstructure targets such as bid-ask spread, trade size, or price impact, where linear baselines already exist.
  • The paper's threshold parameter turns the OFI forecast into an executable rule; a natural next test sweeps the threshold against transaction costs to see whether the accuracy gains translate into net profit.
  • A natural verification step is to hold out entire trading days or assets during the 120-combination hyperparameter search, because those combinations were selected on the same datasets later used for validation.
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 / 4 minor

Summary. The paper proposes a hybrid VAR-FNN model for predicting order flow imbalance (OFI) and a threshold-based trading-intensity signal. A VAR model is fit to buy/sell order counts, a feedforward neural network is trained on the residuals, and the final OFI forecast combines the two components. The model is evaluated on two Binance cryptocurrency datasets (BTCUSD and ETCUSDT) and one synthetic dataset, with comparisons to standalone VAR and FNN models. The abstract and conclusion claim that the hybrid achieves superior forecasting accuracy across all datasets, making it the most effective model for OFI prediction.

Significance. If established, the hybrid's incremental gains over standalone models would be a modest but useful empirical contribution to OFI forecasting. The paper makes some effort toward reproducibility: it reports a sensitivity analysis over 120 hyperparameter configurations, includes time-complexity derivations, and states that code is available (via a placeholder link). However, the evaluation as presented does not support the central claim: the validation data overlap with hyperparameter selection, one result in Table 4 contradicts the claimed universal superiority, the intensity metrics depend on an unreported threshold, and the synthetic dataset is not an independent test. The paper's core result is therefore not currently established, and substantial re-analysis and rewriting are needed.

major comments (5)
  1. [Section 4.2.1 / Table 4] Hyperparameter selection uses the same three datasets that later appear as validation. The text states that 120 parameter combinations were evaluated on the three datasets to select the optimal configuration (lag order 2, 32-16-2, ReLU, Adam), and Table 4 then reports validation metrics on those same three datasets. This is a post-selection evaluation, not an out-of-sample test, so the reported superiority of the hybrid model may reflect overfitting to the validation data rather than genuine predictive skill. A held-out test set or nested cross-validation is required before the claim of superior forecasting accuracy can be accepted.
  2. [Table 4 / Section 4.5] The ETCUSDT row in Table 4 directly contradicts the paper's central claim. The hybrid model has an MAE of 0.031, which is worse than the FNN-only MAE of 0.019, while the abstract and Section 4.5 claim that the hybrid 'consistently outperformed' both alternatives with lower MSE and MAE. This internal inconsistency means the claim of uniform superiority is not supported by the paper's own results, and the summary of findings in Section 4.5 must be revised to acknowledge this exception.
  3. [Section 2 / Table 4] The trading-intensity signal is defined in terms of a threshold T, but no numerical value of T is reported anywhere in the manuscript. Consequently, the 'Accuracy (Intensity)' and 'Precision (Intensity)' columns in Table 4 cannot be interpreted or reproduced, and the intensity-prediction advantage claimed for the hybrid model is not verifiable. The authors must report the threshold used for each dataset and ideally show how the metrics vary with T.
  4. [Section 4.1] The synthetic dataset is described as 'designed to replicate characteristics similar to the real datasets,' but no generative model or parameter values are provided. As a result, the synthetic validation is not an independent test of generalization; it is at best a test on data constructed to resemble the training distribution. The abstract's claim of superior performance on synthetic data is therefore much weaker than stated, and the authors should either provide a detailed synthetic-data generation procedure or drop the synthetic-data claim entirely.
  5. [Appendix A.1] The reported VAR(2) estimation on the training data shows that the coefficients on L1.buy_orders, L1.sell_orders, L2.buy_orders, and L2.sell_orders all have p-values above 0.05 in the buy_orders equation (0.706, 0.268, 0.957, 0.095, respectively), and the sell_orders equation is truncated. This means the VAR component contributes essentially no statistically significant linear explanatory power in that equation, which undermines the paper's mechanistic claim that the hybrid achieves its gains by combining a meaningful linear VAR component with a nonlinear FNN residual stage. The authors should discuss this result explicitly and justify why the hybrid framework is still preferable to an FNN-only model.
minor comments (4)
  1. [Section 4.5 / Figures 5-7] The text refers to 'BTCUSD and ETHUSDT Analysis' and Figures 5-7 are labeled ETHUSDT, but Table 4 provides validation results only for BTCUSD, ETCUSDT, and Synthetic, with no ETHUSDT table. The dataset naming should be made consistent throughout the manuscript.
  2. [Appendix A.1] The regression output for the sell_orders equation is incomplete; only a single coefficient row is shown. The full coefficient table should be included for reproducibility.
  3. [Appendix A.5] The GitHub repository link appears as a placeholder ('GitHub Repository Link') rather than an actual URL, so the claimed access to the source code and the sensitivity-analysis CSV is not verifiable.
  4. [Table 5] Table 5 does not state which time period or sample split it is taken from, so the 'Index' entries cannot be contextualized; a caption should identify the time stamps or sample indices.

Circularity Check

1 steps flagged · score 4.0 of 10

The hybrid VAR-FNN model is a standard residual decomposition and is not circular by construction; however, the validation claim partially reduces to a hyperparameter-selection step because the same three datasets are used both to choose the optimal configuration and to report validation superiority.

  1. fitted input called prediction [Section 4.2.1 (Optimal Configuration) and Section 4.4 (Table 4)]
    "A total of 120 parameter combinations were evaluated ... Model performance was assessed on three datasets using key evaluation metrics: Mean Squared Error (MSE), Mean Absolute Error (MAE), and R2. ... Optimal Configuration: The combination of lag order 2, FNN layer structure 32-16-2, ReLU activation function, and Adam optimizer provided the best results across all datasets ... Table 4 provides the model validation results across the three datasets."

    The three datasets used to select the optimal hyperparameters (BTCUSD, ETCUSDT, Synthetic) are the same three datasets presented as 'validation' in Table 4. The optimal configuration was chosen precisely because it gave the best MSE/MAE/R2 on those datasets, so the Table 4 metrics are post-selection, in-sample values for the hyperparameters. The claimed out-of-sample superiority of the hybrid model is therefore not an independent prediction; it is the result of fitting model configuration to the validation data and then reporting that same fit as validation.

full rationale

The model definition itself is not circular: OFI is defined from buy/sell counts (Eq. 1), VAR forecasts the counts (Eq. 2), residuals are passed to an FNN (Eq. 3), and the final OFI prediction is a genuine additive decomposition rather than a restatement of the target. There is no load-bearing self-citation chain, no imported uniqueness theorem, and no ansatz smuggled in by the authors' prior work. The main circularity concern is in the evaluation protocol: Section 4.2.1 describes selecting the optimal configuration using the same three datasets that Section 4.4 reports as validation, so the validation metrics are contaminated by hyperparameter selection. This does not make the model equations self-referential, but it does mean the central claim of 'superior forecasting accuracy' is not established on independent data. Secondary non-circular concerns include the undisclosed threshold T used for intensity signals, the synthetic dataset being designed to mimic the real datasets, the ETCUSDT row where FNN MAE (0.019) beats hybrid MAE (0.031), and Appendix A.1 showing statistically insignificant VAR coefficients. These are correctness/reproducibility issues, not circularity, and are weighed in the score only insofar as they reinforce that the empirical case is weaker than the text claims.

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

The central claim rests on several unstated or weakly supported premises: the undefined threshold T and window h, the assumption that VAR residuals contain learnable structure, the choice of hyperparameters on the same validation sets, and an undescribed synthetic dataset. These are listed as free parameters and axioms above.

free parameters (5)
  • Intensity threshold T = Not reported
    The BUY/SELL/HOLD signal is defined as OFI > T, OFI < -T, else HOLD, but T is never given. All intensity accuracy and precision figures in Table 4 depend on this choice.
  • OFI window length h = Not reported
    The OFI definition in Eq. (1) depends on a window h over which trades are counted. The value of h used in the experiments is never stated, making the OFI series and all downstream metrics non-reproducible.
  • VAR lag order p = 2
    Selected via sensitivity analysis over {1,2,5,10}. The reported VAR regression in Appendix A.1 shows lagged coefficients that are mostly statistically insignificant, so the 'linear dependencies' the VAR is claimed to capture are not evident in the data.
  • FNN architecture and training hyperparameters = 32-16-2 layers, ReLU, Adam lr=0.001, batch size 8, 50 epochs with early stopping
    Chosen from 120 configurations evaluated on the same three validation datasets. No seed or multiple-run averaging reported, so results may vary.
  • Synthetic data generation parameters = Not described
    The synthetic dataset is said to 'replicate characteristics similar to the real datasets' but the generative model is never described, so it cannot be audited.
assumptions (5)
  • domain assumption OFI, defined by Eq. (1) as (Delta N_B - Delta N_S)/(Delta N_B + Delta N_S), is an appropriate target for prediction.
    The paper assumes this normalized trade-count imbalance is a meaningful, predictable quantity for HFT, but does not validate this against price impact or other external benchmarks.
  • domain assumption The residuals of a linear VAR on buy/sell order counts contain learnable non-linear structure that a feedforward network can capture.
    This is the core motivation for the hybrid, but the paper provides no diagnostic evidence for non-linear residual structure.
  • domain assumption Order counts are well approximated by a VAR model, i.e., linear Gaussian dynamics with constant variance.
    The VAR is fit via OLS; the regression table in Appendix A.1 shows insignificant lag coefficients and an unexplained sell_orders equation, so this premise is questionable.
  • ad hoc to paper The three validation datasets are independent of the hyperparameter selection process.
    Section 4.2.1 states model performance was assessed on these datasets to pick the optimal configuration; the same datasets are then reported as validation results in Table 4, so the evaluation is in-sample for hyperparameters.
  • ad hoc to paper The synthetic dataset is representative of real HFT conditions.
    The synthetic data is generated to resemble the real data, but no description or generative rule is given, so generalization claims rest on an untested assumption.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Hybrid Vector Auto Regression and Neural Network Model for Order Flow Imbalance Prediction in High Frequency Trading." pith.science (2026). https://pith.science/paper/5LCWJKAI

@misc{pith2026241108382,
  author       = {Pith},
  title        = {Pith review of: Hybrid Vector Auto Regression and Neural Network Model for Order Flow Imbalance Prediction in High Frequency Trading},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5LCWJKAI}},
  note         = {Machine review of arXiv:2411.08382}
}
read the original abstract

In high frequency trading, accurate prediction of Order Flow Imbalance (OFI) is crucial for understanding market dynamics and maintaining liquidity. This paper introduces a hybrid predictive model that combines Vector Auto Regression (VAR) with a simple feedforward neural network (FNN) to forecast OFI and assess trading intensity. The VAR component captures linear dependencies, while residuals are fed into the FNN to model non-linear patterns, enabling a comprehensive approach to OFI prediction. Additionally, the model calculates the intensity on the Buy or Sell side, providing insights into which side holds greater trading pressure. These insights facilitate the development of trading strategies by identifying periods of high buy or sell intensity. Using both synthetic and real trading data from Binance, we demonstrate that the hybrid model offers significant improvements in predictive accuracy and enhances strategic decision-making based on OFI dynamics. Furthermore, we compare the hybrid models performance with standalone FNN and VAR models, showing that the hybrid approach achieves superior forecasting accuracy across both synthetic and real datasets, making it the most effective model for OFI prediction in high frequency trading.

Figures

Figures reproduced from arXiv: 2411.08382 by the authors.

Figure 1
Figure 1. Basic architecture of a feedforward neural network (FNN) for financial prediction. [PITH_FULL_IMAGE:figures/full_fig_p005_1.png] view at source ↗
Figure 2
Figure 2. Workflow of the Hybrid VAR-FNN Model for OFI Prediction [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Sensitivity Analysis Heatmaps: Evaluation Metrics Across Parameter Configurations [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Predictions from the Hybrid VAR-FNN Model during Training. [PITH_FULL_IMAGE:figures/full_fig_p011_4.png]
Figure 5
Figure 5. Figure 5: ETHUSDT Actual OFI [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: ETHUSDT Predicted OFI [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]
Figure 7
Figure 7. Figure 7: ETHUSDT Combined Model OFI Prediction Our findings indicate that the hybrid VAR-FNN model effectively leverages the interpretability of traditional econometric time series models with the adaptability of machine learning, making it a robust tool for predicting OFI in v…
Figure 8
Figure 8. Figure 8: shows the loss curve for the standalone FNN model. The graph indicates how the model’s loss decreased over the epochs, along with validation loss to monitor overfitting or underfitting behavior. 2. Training and Validation Loss Curves for Hybrid VAR-FNN Model [PITH_FUL…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

30 extracted references · 20 canonical work pages

  1. [1]

    Cross-impact of order flow imbalance in equity markets

    Rama Cont, Mihai Cucuringu, and Chao Zhang. Cross-impact of order flow imbalance in equity markets. Quantitative Finance, 23(10):1373–1393, 10 2023. ISSN 1469-7688. doi:10.1080/14697688.2023.2236159

    arXiv 2023

  2. [2]

    López de Prado, and Maureen O’Hara

    David Easley, Marcos M. López de Prado, and Maureen O’Hara. Flow Toxicity and Liquidity in a High-frequency World. Review of Financial Studies, 25(5):1457–1493, 5 2012. ISSN 0893-9454. doi:10.1093/rfs/hhs053

    doi:10.1093/rfs/hhs053 2012

  3. [3]

    Kolm, Jeremy Turiel, and Nicholas Westray

    Petter N. Kolm, Jeremy Turiel, and Nicholas Westray. Deep order flow imbalance: Extracting alpha at multiple horizons from the limit order book. Mathematical Finance, 33(4):1044–1081, 10 2023. ISSN 0960-1627. doi:10.1111/mafi.12413

    doi:10.1111/mafi.12413 2023

  4. [4]

    A vector-autoregression analysis of credit and liquidity factor dynamics in US LIBOR and Euribor swap markets

    Bernard Murphy, Finbarr. A vector-autoregression analysis of credit and liquidity factor dynamics in US LIBOR and Euribor swap markets. Journal of Economics and Finance , 36(2):351–370, 4 2012. ISSN 1055-0925. doi:10.1007/s12197-010-9122-2

    work page doi:10.1007/s12197-010-9122-2 2012

  5. [5]

    Estimation of slowly decreasing Hawkes kernels: application to high-frequency order book dynamics

    Emmanuel Bacry, Thibault Jaisson, and Jean–François Muzy. Estimation of slowly decreasing Hawkes kernels: application to high-frequency order book dynamics. Quantitative Finance, 16(8):1179–1201, 8 2016. ISSN 14697696. doi:10.1080/14697688.2015.1123287

    arXiv 2016

  6. [6]

    NEURAL NETWORKS IN FINANCE AND ECONOMICS FORECASTING

    WEI Huang, KIN KEUNG LAI, YOSHITERU NAKAMORI, SHOUY ANG W ANG, and LEAN YU. NEURAL NETWORKS IN FINANCE AND ECONOMICS FORECASTING. International Journal of Information Technology & Decision Making, 06(01):113–140, 3 2007. ISSN 0219-6220. doi:10.1142/S021962200700237X

    work page doi:10.1142/s021962200700237x 2007

  7. [7]

    Mahmoudi, and Javier E

    Mohsen Maleki, Darren Wraith, Mohammad R. Mahmoudi, and Javier E. Contreras-Reyes. Asymmetric heavy- tailed vector auto-regressive processes with application to financial data. Journal of Statistical Computation and Simulation, 90(2):324–340, 1 2020. ISSN 0094-9655. doi:10.1080/00949655.2019.1680675

    arXiv 2020

  8. [8]

    Lee A. Smales. Bond futures and order imbalance. Journal of International Financial Markets, Institutions and Money, 26:113–132, 10 2013. ISSN 10424431. doi:10.1016/j.intfin.2013.05.006

    work page doi:10.1016/j.intfin.2013.05.006 2013

Show all 30 references
  1. [9]

    Trade size, order imbalance, and the volatility–volume relation

    Kalok Chan and Wai-Ming Fong. Trade size, order imbalance, and the volatility–volume relation. Journal of Financial Economics, 57(2):247–273, 8 2000. ISSN 0304405X. doi:10.1016/S0304-405X(00)00057-X

    doi

  2. [10]

    R. Cont, A. Kukanov, and S. Stoikov. The Price Impact of Order Book Events.Journal of Financial Econometrics, 12(1):47–88, 1 2014. ISSN 1479-8409. doi:10.1093/jjfinec/nbt003

    2014 doi

  3. [11]

    Vector Autoregressions

    James H Stock and Mark W Watson. Vector Autoregressions. Journal of Economic Perspectives, 15(4):101–115, 11 2001. ISSN 0895-3309. doi:10.1257/jep.15.4.101

    2001 doi

  4. [12]

    Toda and Peter C

    Hiro Y . Toda and Peter C. B. Phillips. Vector autoregression and causality: a theoretical overview and simulation study. Econometric Reviews, 13(2):259–285, 1 1994. ISSN 0747-4938. doi:10.1080/07474939408800286. 14

    1994 doi

  5. [13]

    Forecasting High Frequency Order Flow Imbalance, 8 2024

    Aditya Nittur Anantha and Shashi Jain. Forecasting High Frequency Order Flow Imbalance, 8 2024. URL https://arxiv.org/abs/2408.03594

    2024 arXiv

  6. [14]

    P. R. Burrell and B. O. Folarin. The impact of neural networks in finance. Neural Computing & Applications, 6 (4):193–200, 12 1997. ISSN 0941-0643. doi:10.1007/BF01501506

    1997 doi

  7. [15]

    Neural networks in business: a survey of applications (1992–1998)

    A Vellido. Neural networks in business: a survey of applications (1992–1998). Expert Systems with Applications, 17(1):51–70, 7 1999. ISSN 09574174. doi:10.1016/S0957-4174(99)00016-0

    1992 doi

  8. [16]

    Enhancing trading strategies with order book signals

    Álvaro Cartea, Ryan Donnelly, and Sebastian Jaimungal. Enhancing trading strategies with order book signals. Applied Mathematical Finance, 25(1):1–35, 1 2018. ISSN 1350-486X. doi:10.1080/1350486X.2018.1434009

    2018

  9. [17]

    Hawkes model for price and trades high-frequency dynamics

    Emmanuel Bacry and Jean-François Muzy. Hawkes model for price and trades high-frequency dynamics. Quantitative Finance, 14(7):1147–1166, 7 2014. ISSN 1469-7688. doi:10.1080/14697688.2014.897000

    2014

  10. [18]

    Capturing the Order Imbalance with Hidden Markov Model: A Case of SET50 and KOSPI50

    Polin Wu and Wasin Siwasarit. Capturing the Order Imbalance with Hidden Markov Model: A Case of SET50 and KOSPI50. Asia-Pacific Financial Markets, 27(1):115–144, 3 2020. ISSN 1387-2834. doi:10.1007/s10690- 019-09285-1

    2020 doi

  11. [19]

    Optimal Execution with Dynamic Order Flow Imbalance

    Kyle Bechler and Michael Ludkovski. Optimal Execution with Dynamic Order Flow Imbalance. SIAM Journal on Financial Mathematics, 6(1):1123–1151, 1 2015. ISSN 1945-497X. doi:10.1137/140992254

    2015 doi

  12. [20]

    Order Imbalance, Liquidity, and Market Efficiency: Evidence from the Chinese Stock Market

    Lei Jiang. Order Imbalance, Liquidity, and Market Efficiency: Evidence from the Chinese Stock Market. Managerial and Decision Economics, 32(7):469–480, 10 2011. ISSN 01436570. doi:10.1002/mde.1547

    2011 doi

  13. [21]

    Fung and Philip L.H

    Joseph K.W. Fung and Philip L.H. Yu. Order imbalance and the dynamics of index and futures prices. Journal of Futures Markets, 27(12):1129–1157, 12 2007. ISSN 0270-7314. doi:10.1002/fut.20288

    2007 doi

  14. [22]

    Effect of order flow imbalance on market impact across market states

    Andrew Kromkowski, Mason Montgomery, Kaustav Saha, Feiyin Wu, and Peter Beling. Effect of order flow imbalance on market impact across market states. In 2016 IEEE Systems and Information Engineering Design Symposium (SIEDS), pages 298–302. IEEE, 4 2016. ISBN 978-1-5090-0970-1....

    2016

  15. [23]

    Dynamic relations between order imbalance, volatility and return of top gainers

    Yong-Chern Su, Han-Ching Huang, and Shiue-Fang Lin. Dynamic relations between order imbalance, volatility and return of top gainers. Applied Economics , 44(12):1509–1519, 4 2012. ISSN 0003-6846. doi:10.1080/00036846.2010.543080. A Appendix A.1 V AR Model Training Summary Summa...

    2012

  16. [24]

    • For lag order p, complexity per variable is O(n · p2)

    V AR Model Training: • The V AR model uses Ordinary Least Squares (OLS) regression for each variable. • For lag order p, complexity per variable is O(n · p2). • With two variables (buy and sell orders), the total complexity becomes: O(2 · n · p2) = O(n · p2)

  17. [25]

    OFI Calculation: • Calculating OFI from V AR-predicted orders has complexityO(n), as it’s an element-wise operation

  18. [26]

    Residual Calculation: • Computing residuals by subtracting V AR predictions from actual orders is an element-wise operation with complexity O(n)

  19. [27]

    • With M epochs, the total complexity for FNN training becomes: O(M · n · d · h)

    FNN Training on Residuals: • For each epoch, the forward and backward passes through the FNN have complexity O(n · d · h). • With M epochs, the total complexity for FNN training becomes: O(M · n · d · h)

  20. [28]

    Combining V AR and FNN-Predicted OFI: • The final OFI calculation requires element-wise addition, with a complexity of O(n). 16 A.3.3 Total Complexity The combined time complexity is: O(n · p2) + O(n) + O(n) + O(M · n · d · h) + O(n) Simplifying, the dominant terms give: O(n ·...

  21. [29]

    The graph indicates how the model’s loss decreased over the epochs, along with validation loss to monitor overfitting or underfitting behavior

    Training and Validation Loss Curves for FNN-Only Model Figure 8 shows the loss curve for the standalone FNN model. The graph indicates how the model’s loss decreased over the epochs, along with validation loss to monitor overfitting or underfitting behavior

  22. [30]

    The steady decline in loss, along with minimal divergence between training and validation loss, suggests effective learning and good generalization on unseen data

    Training and Validation Loss Curves for Hybrid V AR-FNN Model Figure 9 presents the loss curve for the hybrid V AR-FNN model. The steady decline in loss, along with minimal divergence between training and validation loss, suggests effective learning and good generalization on ...

Pith tools

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