Pith. sign in

REVIEW 5 major objections 5 minor 4 cited by

Tail Risk Alert Based on Conditional Autoregressive VaR by Regression Quantiles and Machine Learning Algorithms

T0 review · 5 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read The US credit bond market is the main transmitter of tail risk to other US markets, with larger and longer-lasting spillovers than any reverse effect.

desk verdict Applies a known MVMQ-CAViaR framework to US data and finds credit-market tail-risk centrality, but the predictive claim rests entirely on in-sample fits and an undefined pseudo-IRF. read the letter →

arxiv 2412.06193 v2 pith:743DQ4RO submitted 2024-12-09 q-fin.RM

Zong Ke , Yuchen Yin This is my paper · ORCID
classification q-fin.RM
keywords tailriskspilloverCAViaRcreditbondmarketearlywarninggeneticalgorithmgradientdescentValue-at-Riskquantileregression
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 aims to establish that the US credit bond market, not the stock market, is the main origin of extreme (tail) risk for the other major US financial markets. Using a multivariate CAViaR model that models the $1\%$ Value-at-Risk of each market, the paper estimates how a tail event in one market shifts the tail risk of another. It reports that credit-market tail shocks spill into the stock market with greater size and persistence than any spillover into the credit market, while foreign-exchange and interbank shocks barely affect credit. If this asymmetry is correct, monitoring high-yield credit spreads could provide an early warning of stock-market crashes and improve systemic risk management.

What carries the argument

The main object is the MVMQ-CAViaR model (multivariate multi-quantile conditional autoregressive Value-at-Risk): a two-equation system in which the $1\%$ VaR of each market depends on its own lagged VaR, the other market's lagged VaR, and lagged shocks. The off-diagonal coefficients $b_{12}/b_{21}$ and $a_{12}/a_{21}$ measure directional tail-risk spillovers, and joint tests on those coefficients decide whether a market transmits, receives, or both. Model parameters are estimated by regression quantiles with gradient descent and genetic algorithm optimization, with the genetic algorithm reported as the better optimizer.

What would settle it

Re-estimate the model on data through 2019 and generate one-step-ahead $1\%$ VaR forecasts for the S&P 500 for 2020–2024 from the credit-market equations; if these forecasts do not beat a univariate CAViaR benchmark on quantile scores or hit rates, the credit-market-centrality claim loses its predictive content.

Watch

Extended reading notes

Core claim

The central claim is that the credit bond market is a systemic risk hub: its extreme risk events transmit to the stock, foreign exchange, and interbank markets, while only stock-market tail risk feeds back into credit. This is established through the MVMQ-CAViaR estimates, joint hypothesis tests on the spillover coefficients, and pseudo-impulse response functions, which together show the credit-to-stock spillover is larger and longer-lived than the stock-to-credit spillover. The paper states that historical extreme-risk information from the credit bond market can therefore serve as a predictor of the Value-at-Risk of other markets and places credit in a central warning position.

Load-bearing premise

The central result assumes that the spillover coefficients estimated on the full 2014–2024 sample are stable and capture genuine one-way causation, so that the fitted model's impulse responses describe how future tail risks will actually propagate; no out-of-sample forecasting evaluation is reported.

Editorial extensions

If this is right

  • Regulators and investors should treat the US high-yield credit market as a leading indicator for equity tail risk.
  • Stock portfolio risk models should include lagged credit-spread quantiles as conditioning variables.
  • Tail shocks from the foreign-exchange and interbank markets have comparatively little effect on credit-bond risk, so early-warning systems can focus on credit and equity.
  • The significant two-way spillover between stocks and credit implies that tail-risk hedging across these two markets is more valuable than hedging credit against FX or interbank risk.

Reading between the lines

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

  • The paper's pseudo-impulse responses are computed in-sample; a true out-of-sample forecasting exercise would be needed to turn the credit-market-centrality result into a usable early-warning system.
  • The reported asymmetry is consistent with structural credit-risk models in which default risk drives equity values, but the paper does not test that underlying mechanism.
  • Using HIBOR, a Hong Kong rate, as the US interbank proxy could weaken the interbank results; substituting a US-based rate such as SOFR might change the estimated spillover network.
  • If the asymmetry holds, it implies an asymmetric information flow in which credit markets aggregate distress information earlier than equity markets, a prediction that could be tested with Granger causality in tail quantiles.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

5 major / 5 minor

Summary. The manuscript estimates bivariate multivariate-quantile CAViaR systems at the 1% quantile for three pairs of markets—credit bonds with stocks, with foreign exchange, and with interbank lending—using daily data from May 2014 to June 2024. Parameters are estimated by gradient descent and genetic algorithm, and the paper uses joint significance tests and a figure called "pseudo-impulse response" to claim that the credit bond market is the dominant source of tail-risk spillovers and that its historical tail-risk information can predict other markets' VaR. The central claims are that the credit market occupies a "central warning position" and that its extreme-risk history serves as a predictor for stocks, FX, and interbank markets.

Significance. The question of which US market leads tail-risk transmission is practically important for systemic-risk monitoring, and the MVMQ-CAViaR framework is a reasonable starting point for studying quantile spillovers. The paper's merits include the use of a recognized conditional autoregressive quantile model, a long daily sample, and a transparent comparison of two optimization approaches. However, the significance of the contribution depends on the early-warning claim, and that claim is not established by the evidence as presented. The missing out-of-sample validation, the undefined pseudo-impulse-response procedure, and the apparent misreporting of significance tests are all load-bearing problems.

major comments (5)
  1. [§III.C, Figure 5] The pseudo-impulse response procedure used for Figure 5 is never defined. The paper does not specify how a one-unit shock to one market's VaR is imposed in the nonlinear MVMQ-CAViaR system, how many horizons are traced, whether coefficients are held fixed, or how the duration comparison is measured. Because the conclusion that the credit market's spillover effect and duration are largest is read directly from this figure, the missing definition is a load-bearing gap.
  2. [Abstract and §IV] The claim that historical extreme-risk information from the credit market can 'serve as a predictor' of other markets' VaR is not supported by any out-of-sample exercise. All coefficients in Equations (1)-(2) and Table 3 are estimated on the full 2014-2024 sample, and the joint tests and Figure 5 are in-sample summaries. No holdout period, rolling-window forecast, forecast-encompassing test, or VaR exceedance backtest is reported. Given the autoregressive terms b11 and b22, significant cross-coefficients do not by themselves imply predictive gains over a univariate CAViaR benchmark.
  3. [Table 5] The column heading 'Accept/Refuse' and the entry 'refuse' are incorrect for p-values of 0.0619, 0.4906, and 0.4954. At the conventional 5% level these null hypotheses cannot be rejected, and the latter two cannot be rejected at the 10% level either. This is not a labeling detail: the directional conclusion that FX and interbank markets are spillover receivers in Models 1 and 3 depends on rejecting a12=b12=0, while the corresponding a21=b21=0 rows actually show no significant reverse spillover.
  4. [Table 3 and §III.B.2] The text states that in Model 2 (stock/credit) b12 and b21 are both significant at the 5% level. The displayed estimate for b21 is -0.0103 with a standard error of 0.0117, which gives a t-ratio of about -0.88 and is not significant. The asterisk alignment in Table 3 is also unclear, with parentheses and stars seemingly attached to the wrong coefficients in several rows, so the significance claims in this table cannot be verified as printed.
  5. [Table 1 and §III.B.1] HIBOR, a Hong Kong interbank rate, is used as the proxy for the US interbank market. Since the paper's stated object is US financial markets, the interbank-market results and Figure 5's claim about 'the other two markets' are only valid if HIBOR tracks US interbank conditions; no justification or sensitivity analysis with a US-based rate (for example, SOFR or the TED spread) is provided.
minor comments (5)
  1. [§III.B.2] The subsection labels (a)-(c) are inconsistent with Table 3: the text says 'Model 3 reports risk spillovers in the foreign exchange market and credit bond market,' but Table 3 lists Model 3 as interbank/credit and Model 1 as FX/credit, which makes the estimation results hard to follow.
  2. [Equation (7)] The quantile regression objective is written without the usual check function and the notation is garbled; please restate it as min over beta of (1/n) times the sum of rho_k(r_t - f_t(beta)).
  3. [§III.B.3, Figures 2-4] The comparison between genetic algorithm and gradient descent is informal; no hyperparameters (mutation rate, crossover rate, learning-rate schedule) or convergence criteria are reported, and it is unclear whether the loss curves shown are the same objective function used in the CAViaR estimation.
  4. [References] Several citations do not correspond to the claimed content (for example, [12] is cited for credit risk in China but is a deep-network pattern recognition paper), and the reference list includes items never discussed in the text, such as [9] and [16].
  5. [Table 3 title] The table title reads 'MCMQ-CAViaR' but the model is elsewhere called MVMQ-CAViaR; the typo should be corrected.

Circularity Check

1 steps flagged · score 6.0 of 10

The credit market's 'early warning' and 'predictor' claims are based on in-sample fitted conditional quantiles, with no out-of-sample validation; the predictive conclusion reduces to the fitted model by construction.

  1. fitted input called prediction [Abstract; Section III.B.2 (Model estimations, Table 3); Section III.C(2) (Pseudo-impulse response, Figure 5); Conclusions]
    "The model is used to provide early warning of related risks in US stocks, US credit bonds, etc. The results show that, by analyzing the direction, magnitude, and pseudo-impulse response of the risk spillover, it is found that the credit market's spillover effect on the stock market and its duration are both greater than the spillover effect of the stock market and the other two markets on credit market, placing credit market in a central position for warning of extreme risks. Its historical information on extreme risks can serve as a predictor of the VaR of other markets."

    The only evidence offered for the early-warning/predictor claim is the full-sample MVMQ-CAViaR estimation over 2014-2024 (Section III.B.2, Table 3) and the pseudo-impulse responses in Figure 5, both computed from the same fitted coefficients of Equations (1)-(2). Those equations specify the target market's conditional quantile as a function of lagged RCB tail risk whenever the cross-coefficient is nonzero, so a significant coefficient is an in-sample property of the fitted quantile, not an out-of-sample predictive validation. No held-out period, rolling-window forecast, structural-break test, exceedance evaluation, or univariate CAViaR benchmark is reported. Thus 'credit market historical information can predict VaR of other markets' is a fitted input relabeled as a prediction.

full rationale

The central predictive claim in the abstract and conclusions — that the credit bond market's historical extreme-risk information 'can serve as a predictor of the VaR of other markets' and that it occupies a 'central warning position' — is supported only by the full-sample fit of Equations (1)-(2), Table 3, and by pseudo-impulse responses generated from that same fit. Because lagged RCB tail risk enters the other markets' conditional quantile equations by construction, the significant cross-coefficients are an in-sample description of the fitted quantiles. No out-of-sample forecast exercise or comparison against external benchmarks is reported, so the 'predictor' terminology converts a fitted model into a claimed predictive fact. The remainder of the spillover analysis — significant cross-coefficients, joint hypothesis tests, and pseudo-impulse-response shapes — is a legitimate historical description of the fitted sample, but the paper's own framing makes the predictive conclusion reduce, by construction, to the fitted inputs. Self-citations in the reference list are not load-bearing and do not add circularity beyond this step.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The central claim rests on a fitted linear quantile autoregression with ten coefficients per market pair, several hand-chosen proxies and hyperparameters, and no out-of-sample validation. The model structure is taken from the CAViaR literature, so the main burden is on the stability of the estimated coefficients and the appropriateness of the proxies. No new entities are introduced.

free parameters (5)
  • MVMQ-CAViaR coefficients (c, a, b matrices) = Table 3, ten coefficients per pairwise model
    All coefficients are fitted to the 2014-2024 daily return data via gradient descent and genetic algorithm; the spillover and 'centrality' conclusions are read directly off these fitted values.
  • Quantile level tau = 1%
    The 1% quantile is chosen by the authors; results would likely differ at other quantile levels and no sensitivity analysis is reported.
  • Genetic algorithm hyperparameters = Population 500, iterations 1000; other hyperparameters unspecified
    Population size and iteration count are chosen by hand; crossover and mutation rates are not stated, and no sensitivity analysis is shown.
  • Gradient descent learning rate = r in Eq (4), value not reported
    The learning rate is a hand-chosen tuning parameter; the convergence claims depend on it.
  • Proxy variable selection = ICE BofA US High Yield Effective Yield, SPX, HIBOR, USD Index
    The choice of proxies is an input; using HIBOR, a Hong Kong rate, for the US interbank market is especially questionable.
assumptions (4)
  • domain assumption CAViaR quantile dynamics are correctly specified as linear autoregressions in lagged returns and lagged VaRs (Eqs 1-2).
    The central spillover estimates are only meaningful if the MVMQ-CAViaR specification is the true data-generating process for the 1% conditional quantiles.
  • ad hoc to paper The optimization algorithms converge to the global minimum of the quantile regression objective.
    No convergence proof or multiple-restart analysis is given; the paper asserts convergence from Figures 2 and 3.
  • domain assumption Return series are stationary over May 2014 to June 2024 with no structural breaks.
    ADF tests are reported, but the model ignores regime shifts such as COVID-19 and the 2022 rate cycle, which could change spillover direction.
  • domain assumption The proxy variables represent the intended markets.
    HIBOR is a Hong Kong rate, not a US interbank rate; ICE BofA High Yield Effective Yield is a yield level rather than a return.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Tail Risk Alert Based on Conditional Autoregressive VaR by Regression Quantiles and Machine Learning Algorithms." pith.science (2026). https://pith.science/paper/743DQ4RO

@misc{pith2026241206193,
  author       = {Pith},
  title        = {Pith review of: Tail Risk Alert Based on Conditional Autoregressive VaR by Regression Quantiles and Machine Learning Algorithms},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/743DQ4RO}},
  note         = {Machine review of arXiv:2412.06193}
}
read the original abstract

As the increasing application of AI in finance, this paper will leverage AI algorithms to examine tail risk and develop a model to alter tail risk to promote the stability of US financial markets, and enhance the resilience of the US economy. Specifically, the paper constructs a multivariate multilevel CAViaR model, optimized by gradient descent and genetic algorithm, to study the tail risk spillover between the US stock market, foreign exchange market and credit market. The model is used to provide early warning of related risks in US stocks, US credit bonds, etc. The results show that, by analyzing the direction, magnitude, and pseudo-impulse response of the risk spillover, it is found that the credit market's spillover effect on the stock market and its duration are both greater than the spillover effect of the stock market and the other two markets on credit market, placing credit market in a central position for warning of extreme risks. Its historical information on extreme risks can serve as a predictor of the VaR of other markets.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 4 Pith papers

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

  1. HADES: Hardware Accelerated Decoding for Efficient Speculation in Large Language Models

    cs.CL 2024-12 conditional novelty 5.0 of 10

    A custom Verilog module accelerates the token-acceptance step of speculative decoding by about 7x over GPUs, but the step is only a small fraction of total LLM inference.

  2. Developing Cryptocurrency Trading Strategy Based on Autoencoder-CNN-GANs Algorithms

    cs.LG 2024-12 reject novelty 4.0 of 10

    A standard autoencoder-CNN-GAN stack is applied to Bitcoin futures price prediction, with reported accuracy and profits that the paper does not adequately support.

  3. Scam Detection for Ethereum Smart Contracts: Leveraging Graph Representation Learning for Secure Blockchain

    cs.LG 2024-12 reject novelty 3.0 of 10

    A simple neural net trained on aggregated graph features of Ethereum transactions claims higher scam-detection F1 than a graph neural network, but the tiny, synthetic test set makes the comparison unreliable.

  4. Regression and Forecasting of U.S. Stock Returns Based on LSTM

    q-fin.ST 2025-02 reject novelty 2.0 of 10

    The authors fit standard factor models and an LSTM to U.S. sector returns and report that the five-factor model and LSTM each look best in different sectors.

Reference graph

Works this paper leans on

33 extracted references · 31 canonical work pages · cited by 4 Pith papers

  1. [1]

    Xue, J, Dai, X., Zhang, D., Xuan-Hoa Nghiem, Wang, Q. (2024). Tail risk spillover network among green bond, energy and agricultural markets under extreme weather scenarios, International Review of Economics&Finance,103707

    work page 2024

  2. [2]

    Dahl, R.E., Oglend, A., & Yahya, M. (2020). Dynamics of volatility spilloverincommoditymarkets:Linkingcrudeoiltoagriculture.Journal ofCommodityMarkets,20,100111

    work page 2020

  3. [3]

    Li,K.,Xie,C.,Ouyang,Y.,Mo,T.,Feng,Y.(2024).Tailriskspillovers inthe stock and forex markets at the major emergencies: Evidence from the G20 countries, International Review of Financial Analysis, Volume 96,PartB,103712

    work page 2024

  4. [4]

    Yu, P., Zhou, L., Chen, Z., Li, C. (2024). Risk spillover changes among commodity futures, stock and ESG markets: A study based on multidimensional higher order moment perspective, Finance Research Letters,106284

    work page 2024

  5. [5]

    Gnagne, P.X., Simo-Kengne, B.D., Manguzvane, M.M. (2024). The spillover and contagion effects of sovereign risk on stock markets, EconomicModelling,Volume141,106921

    work page 2024

  6. [6]

    Qu, T. (2024). Tail-driven portfolios: Unveiling financial contagion and enhancing risk management, The Journal of Finance and Data Science, Volume10,100142

    work page 2024

  7. [7]

    Kal, S.H., Arslaner, F., Arslaner, N. (2015). The dynamic relationship between stock, bond and foreign exchange markets, Economic Systems, Volume39,Issue4,Pages592-607

    work page 2015

  8. [8]

    Ehlers,P.,&Schönbucher,P.(2006).TheinfluenceofFXriskoncredit spreads.WP,ETHZurich

    work page 2006

Show all 33 references
  1. [9]

    Zhao, P., Fan, R., Wang, S., Shen, L., Zhang, Q., Ke, Z., & Zheng, T. (2024). Contextual bandits for unbounded context distributions. arXiv preprintarXiv:2408.09655

    2024 arXiv

  2. [10]

    Schonbucher, P., & Ehlers, P. (2014). The Influence of FX Risk on CreditSpreads,UniversityofZürich,NCCRFINRISK:214

    2014

  3. [11]

    Pierre, C.D., Bjerre, T.A., Benjamin, J. (2023). How Integrated Are CreditandEquityMarkets?TheJournalofFinance,Volume79,Issue2: 949-992

    2023

  4. [12]

    Gao, Z., Wu, Y., Bu, X., Yu, T., Yuan, J., & Jia, Y. (2019). Learning a robust representation via a deep network on symmetric positive definite manifolds.PatternRecognition,92,1-12

    2019

  5. [13]

    Zhou,F.(2022).RiskAlertofCreditmarkets,ChinaPrice,No4:29-33

    2022

  6. [14]

    Juan Andres Rodriguez-Nieto and Andre V. Mollick. (2021). Financial MarketsandPortfolioManagement,volume35,issue2,No3,225-254

    2021

  7. [15]

    Hofmann, Boris, Shin, Song, H., & Villamizar-Villegas, Mauricio. (2019). FX intervention and domestic credit: Evidence from high- frequency micro data, No. 774, BIS Working Papers, Bank for InternationalSettlements

    2019

  8. [16]

    Large- Scale Object Detection in the Wild from Imbalanced Multi-Labels,

    J. Peng, X. Bu, M. Sun, Z. Zhang, T. Tan and J. Yan. (2020). "Large- Scale Object Detection in the Wild from Imbalanced Multi-Labels," 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR),Seattle,WA,USA,pp.9706-9715

    2020

  9. [17]

    (2024).Incorporating economic indicators andmarketsentimenteffectintoUSTreasurybondyieldpredictionwith machine learning

    Li, Z., Wang, B., &Chen, Y. (2024).Incorporating economic indicators andmarketsentimenteffectintoUSTreasurybondyieldpredictionwith machine learning. Journal of Infrastructure, Policy and Development, 8(9),7671

    2024

  10. [18]

    Bu, X., Wu, Y., Gao, Z., & Jia, Y. (2019). Deep convolutional network with locality and sparsity constraints for texture classification. Pattern Recognition,91,34-46

    2019

  11. [19]

    Wu, W. (2024). Alphanetv4: Alpha Mining Model. 10.48550/arXiv.2411.04409

    2024 doi arXiv

  12. [20]

    Luo, D. (2024). International Journal of Research Publication and Reviews, Enhancing Smart Grid Efficiency through Multi-Agent Systems: A Machine Learning Approach for Optimal Decision Making. 115-120

    2024

  13. [21]

    Wang, B., Chen, Y., & Li, Z. (2024). A novel Bayesian Pay-As-You- Drive insurance model with risk prediction and causal mapping. DecisionAnalyticsJournal,13,100522

    2024

  14. [22]

    Engle, R.F., Manganelli, S., Buchinsky, M., Haan, W.D., Horowitz, J., &Powell, J.(1999).CAViaR:ConditionalAutoregressiveValueatRisk byRegressionQuantiles

    1999

  15. [23]

    Yun, D. (2024). MeGA: Merging Multiple Independently Trained NeuralNetworks Based onGenetic Algorithm,Neural and Evolutionary Computing,vol7,pp:32-46

    2024

  16. [24]

    Zhao, P., & Wan, Z. (2023). High dimensional distributed gradient descent with arbitrary number of byzantine attackers. arXiv preprint arXiv:2307.13352

    2023

  17. [25]

    Ahmed, W. M. A. (2024). Dynamic interactions between Egyptian equityandcurrencymarkets priortoandduringpoliticalunrest,Applied FinancialEconomics,24(20):1347–1359

    2024

  18. [26]

    A. G. Carlon, Torii, A. J., Lopez, R. H., & de Cursi, J. E. S.. (2022). Stochastic Gradient Descent for Risk Optimization, Proceedings of the 5th International Symposium on Uncertainty Quantification and StochasticModelling,424–435

    2022

  19. [27]

    Chen, X. (2022). Dimension independent excess risk by stochastic gradientdescent,ElectronicJournalofStatistics,Vol.16: 4547–4603

    2022

  20. [28]

    Christiansen, C. (2010). Decomposing European bond and equity volatility.Internationaljournaloffinance&economics,15(2),105-122

    2010

  21. [29]

    Ouyang, K., & Fu, S. (2024). Graph neural networks are evolutionary algorithms.arXive-prints,arXiv-2412

    2024

  22. [30]

    Katoch, S., Chauhan,S.S., &VijayKumar. (2021). Areview ongenetic algorithm:past,present,andfuture,volume80:8091–8126

    2021

  23. [31]

    J M García1, C A Acosta1, and M J Mesa. (2020). Genetic algorithms formathematicaloptimization,series1448:36-42

    2020

  24. [32]

    Mirjalili, S. (2019). Genetic Algorithm, Evolutionary Algorithms and NeuralNetworks,2019,vol780:67-98

    2019

  25. [33]

    Velho, G., Auriol, J., Bonalli, R. (2023). A Gradient Descent-Ascent Method for Continuous-Time Risk-Averse Optimal Control, Systems andControlLetters,6:31-49

    2023

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.