{"id":"a8149951-73f7-45ee-b3ca-56e27394eb88","arxiv_id":"1908.05419","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":5,"one_line_summary":"The paper applies standard GARCH, portfolio optimization, and Esscher-transform option pricing to seven crypto assets, but its own results contradict the abstract's claim that both optimized portfolios beat the S&P 500.","lead":"The paper fits seven cryptocurrencies with an ARMA-GARCH model, builds minimum-risk portfolios, and prices hypothetical options on those portfolios using a calibrated NIG model and the Esscher transform. The headline claim that these crypto portfolios beat the S&P 500 is only partly supported: the minimum-CVaR portfolio ended higher, while the minimum-variance portfolio underperformed throughout the sample.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The paper's headline claim is contradicted by its own results: Section 3.1 states the min-variance portfolio stayed below SPY throughout the out-of-sample window, and Table 4 shows it has lower Sharpe and higher drawdown than SPY.","rationale":"The reader's rationale already identifies the internal contradiction between the abstract and Section 3.1/Figure 2, and my stress test confirms it as the single most load-bearing concern. The abstract's claim about 'portfolios' is plural, yet the paper's own text and Table 4 show the min-variance portfolio underperforming SPY on cumulative return, Sharpe ratio, MDD, and Rachev ratio in the stated out-of-sample window. This is not a subtle econometric issue or a matter of disagreement with current consensus; it is an inconsistency within the paper itself, so the advertised central claim cannot stand as written. I did not choose the transaction-cost critique or the same-sample model-selection critique as the primary concern, though both are secondary weaknesses and would matter for any revised 'min-CVaR outperforms' claim. The paper is not without merit: the multivariate GARCH backtesting, risk budgeting, and Esscher-transform machinery are competently assembled, and the finding could be salvaged by correcting the overclaim, adding cost sensitivity, and treating the option prices as illustrative. But because the central claim fails on the paper's own evidence, the existing REJECT verdict remains appropriate without modification.","tokens_in":10516,"tokens_out":4610,"duration_ms":46044,"concrete_test":"Reproduce the rolling-window backtest from Section 3.1 using the stated ARMA(1,1)-GARCH(1,1) model with multivariate t innovations and the min-variance weights, then compute the full-sample cumulative return and the four risk-adjusted performance measures in Table 4 for both the min-variance portfolio and SPY over 04/04/2018-07/02/2019. If the min-variance cumulative return is below SPY for every date, or its Sharpe ratio and Rachev ratio are below SPY's, the abstract's joint outperformance claim is refuted by the paper's own results. As a separate check on the remaining claim, rerun the min-CVaR strategy with realistic crypto transaction costs (e.g., 50-100 basis points per side, applied to daily weight turnover) to see whether its outperformance survives frictions.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract's central claim—that portfolios constructed via optimizations that minimize variance and CVaR outperform the S&P 500—is contradicted by the manuscript's own empirical section. Section 3.1 states: 'The cumulative returns from the min Variance portfolio is very stable and has relatively low cumulative returns than the benchmark (SPY) all the time in the estimation window.' Table 4 confirms: min-variance Sharpe 0.0078 vs SPY 0.0333; MDD 0.7464 vs 0.1935; Rachev ratio 1.0124 vs 1.0246; only M2 is marginally higher (0.0004 vs 0.0003). Thus the min-variance portfolio does not outperform SPY on virtually any reported metric in the only out-of-sample period. The Section 3.1 hedge 'often outperforms' is unsupported by a single one-year window in which the outperformance appears only for min-CVaR and only in part of 2019. Since the paper's main advertised finding fails for one of the two portfolios, the empirical horse race cannot support the headline claim. This is an internal inconsistency, not a matter of consensus: the evidence presented by the authors themselves rules out the abstract's claim. A revision could rephrase the finding as 'min-CVaR shows higher risk-adjusted return than SPY in 04/2018-07/2019 under zero-cost rebalancing,' but the current central claim overstates it.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper models daily log returns of seven crypto assets (BTC, ETH, XRP, LTC, BCH, EOS, BNB) using ARMA(1,1)-GARCH(1,1) dynamics with alternative innovation distributions (Gaussian, Student-t) and multivariate dependence structures (t copula, multivariate t, multivariate variance-gamma). Based on VaR/CVaR backtesting over 04/04/2018-07/02/2019, the authors select ARMA(1,1)-GARCH(1,1) with Gaussian innovations and a multivariate t distribution with five degrees of freedom. They then construct rolling min-variance and min-CVaR portfolios of the seven crypto assets and compare their performance with the SPY over the same window, reporting Sharpe, M2, Rachev ratios, and maximum drawdown. Finally, they fit NIG innovations to the min-variance and min-CVaR portfolio returns and price European options via the Esscher transform, asserting that the resulting theoretical prices would be followed closely by future crypto index options.","tokens_in":10808,"tokens_out":6317,"duration_ms":56917,"significance":"If the empirical claims held, the paper would offer a practical framework for constructing minimum-risk crypto portfolios and a first approach to pricing options on crypto indices. The use of standard GARCH-type dynamics, risk budgeting, and the Esscher transform is competent and the code/data description is mostly reproducible. However, the central advertised finding is contradicted by the paper's own results: the min-variance portfolio underperforms SPY throughout the evaluation window on almost all reported metrics. In addition, the model is selected and evaluated on the same out-of-sample window, and the option pricing section contains no external validation. The paper therefore does not currently support its main conclusions.","major_comments":[{"comment":"The headline claim that 'crypto portfolios constructed via optimizations that minimize variance and Conditional Value at Risk outperform a major stock market index' is contradicted by the authors' own empirical results. Section 3.1 states that 'the cumulative returns from the min Variance portfolio ... has relatively low cumulative returns than the benchmark (SPY) all the time in the estimation window,' and Figure 2 shows the min-variance portfolio below SPY for the entire window. Table 4 reports, for the min-variance portfolio, a Sharpe ratio of 0.0078 versus 0.0333 for SPY, a maximum drawdown of 0.7464 versus 0.1935, and a Rachev ratio of 1.0124 versus 1.0246; only the M2 ratio (0.0004 vs 0.0003) is slightly higher. Thus the advertised outperformance holds only for the min-CVaR portfolio, and only during part of 2019. The abstract and the conclusion in Section 5 must be revised to state the result accurately.","section":"Abstract, Section 3.1, Table 4"},{"comment":"The model is selected and evaluated on the same out-of-sample window. The backtesting in Table 2 uses data from 04/04/2018 to 07/02/2019 to choose the multivariate t innovation model, and Table 4 reports portfolio performance over exactly the same 455-day window. Consequently, the reported 'out-of-sample' Sharpe, M2, and Rachev ratios are not independent of the model choice; there is no hold-out period or cross-validation. This circularity weakens the claim that the model was 'carefully backtested' and that the portfolio horse race is out-of-sample.","section":"Section 2.2, Tables 2 and 4"},{"comment":"The portfolio comparison assumes no transaction costs, as stated in Section 3.1: 'we assume there are no transaction costs so that weights can be adjusted purely for hedging risk.' The weight plots in Figures 3 and 4 show large and frequent rebalancing, particularly for the min-CVaR portfolio. Given the wide bid-ask spreads observed in crypto markets, the reported performance advantage in Table 4 may vanish once realistic frictions are accounted for. The paper should report portfolio turnover and a cost-sensitivity analysis before claiming outperformance.","section":"Section 3.1, Figures 3 and 4"},{"comment":"The option pricing claim is not supported by the evidence presented. The NIG parameters are estimated from the same 455 historical portfolio returns used to construct the portfolios, and the simulated prices in Equations (28)-(29) are not compared with any external market data because no such options are traded. The abstract's assertion that the model 'was carefully backtested' refers to the VaR/CVaR backtest of the return model in Section 2.2, not to option prices; the statement in Section 4 that future crypto options 'should follow closely our theoretical prices' is therefore an extrapolation without validation. At a minimum, the paper should report Monte Carlo standard errors, sensitivity to the NIG parameter estimates, and a clear caveat that the prices are untested predictions.","section":"Section 4, Equations (26)-(29), Abstract"}],"minor_comments":[{"comment":"Both figures are captioned 'Option Prices based on the min CVaR portfolio,' but the text states that Figure 5 is based on the min Variance portfolio and Figure 6 on the min CVaR portfolio; the captions should be corrected.","section":"Figures 5 and 6"},{"comment":"The notation is confusing: portfolio weights ωt(j) are introduced for the optimized portfolios, but the risk-budgeting analysis then redefines them as equal weights (1/d, ..., 1/d). The authors should clarify that the risk-contribution analysis in Figures 3 and 4 applies to the equally weighted portfolio, not to the optimized min-variance or min-CVaR portfolios.","section":"Section 3.2"},{"comment":"Step 4(b) says 'Renew β as β + √σtθt,' while the Esscher condition in Equation (26) uses β + θ; the scaling with √σt should be explained or corrected, since it affects the risk-neutral innovation distribution.","section":"Section 4, step 4(b)"},{"comment":"The table would benefit from a note specifying the confidence level, the definition of 'observations,' and the exact form of the binomial test used, since the traffic-light and binomial outcomes are central to model selection.","section":"Table 2"},{"comment":"The sentence 'The cumulative returns from the min Variance portfolio is very stable and has relatively low cumulative returns than the benchmark' contains a grammatical error and should read 'lower cumulative returns than the benchmark.'","section":"Section 3.1"}],"recommendation":"reject","confidential_remarks":"The central empirical claim is contradicted by the paper's own tables and figures, and the option-pricing section cannot be validated with existing data. A revision would require a properly designed out-of-sample evaluation, transaction-cost analysis, and a substantially more cautious framing of the option-pricing predictions; in its current form the paper does not meet the standard for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear X,\n\nYou asked about Hu, Rachev, Fabozzi. The paper is a competent application of standard financial econometrics to crypto assets: ARMA(1,1)-GARCH(1,1), multivariate t, Markowitz/CVaR optimization, risk budgeting, and NIG-Esscher option pricing. What's actually new is the specific empirical exercise—seven crypto assets, min-variance/min-CVaR portfolios versus SPY, and hypothetical option smiles for a nonexistent crypto index. The machinery is assembled cleanly, and the backtesting section is thorough (VaR/CVaR failures, traffic-light, binomial tests).\n\nThe soft spots are real, and one is load-bearing. The abstract says both optimal portfolios outperform SPY, but Figure 2 and Table 4 show the min-variance portfolio below SPY for the entire out-of-sample window with lower Sharpe (0.0078 vs 0.0333) and much deeper drawdown. The authors even state this in Section 3.1, then soften to 'often outperforms,' which no metric in the paper supports. Only the min-CVaR portfolio beats SPY, and only from mid-2019. That overclaim needs to be cut.\n\nNext, the model selection and the performance evaluation share the same 04/2018–07/2019 window, so the 'out-of-sample' horse race is not independent of model choice. No confidence intervals or robustness checks around Sharpe or CVaR. And the zero-transaction-cost assumption (Section 3.1) matters: min-CVaR weights swing sharply, and crypto spreads are wide, so a cost-sensitivity analysis could erase the edge entirely.\n\nThe option pricing section is the weakest. The NIG parameters are fitted to the same historical portfolio returns, and there are no traded options to validate against. The claim that future crypto options 'should follow closely our theoretical prices' is calibrated extrapolation, not prediction. It would be fine as an illustrative exercise; it is not a fair-valuation guarantee.\n\nProportionately: this is not a bad paper. The econometrics is standard but honestly executed, and the risk-budgeting results (BTC as diversifier, EOS as contributor) are plausible. With the overclaim removed, the option prices relabeled as illustrative, and a cost/turnover analysis added, this would be a modest empirical contribution to crypto portfolio management. As is, it needs major revision.\n\nMy recommendation: if it comes to you as editor, I would not desk-reject it outright—send it to a referee willing to enforce the reframing. But I would not cite it in its current form.\n\nBest","headline":"A competent but overclaimed application of standard econometric machinery to crypto portfolios; the headline outperformance result contradicts the paper's own tables.","tokens_in":11370,"tokens_out":3181,"would_cite":false,"duration_ms":28431,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91G20","91G10"],"pacs":[],"model":"deepseek-v4-flash","headline":"Minimum-risk portfolios of crypto assets can beat the S&P 500, the paper finds, and the same model prices European options on them.","keywords":["crypto assets","minimum-variance portfolio","conditional value at risk","ARMA-GARCH","normal inverse Gaussian distribution","Esscher transform","crypto option pricing","risk budgeting"],"falsifier":"Re-run the same rolling optimization on data after July 2019 (or between 2017 and 2018) and record the min-CVaR portfolio's cumulative return net of realistic crypto bid-ask spreads and rebalancing fees; if it no longer exceeds the S&P 500 on Sharpe or cumulative return, the paper's major finding is specific to the original window and frictionless assumption.","tokens_in":10285,"feed_emoji":"📈","tokens_out":8913,"duration_ms":78908,"temperature":0.7,"pith_summary":"Using daily returns on the top seven crypto assets from mid-2017 through mid-2019, the paper fits a multivariate ARMA(1,1)-GARCH(1,1) model with Student-t innovations, selects the specification by Monte Carlo backtesting of value at risk and conditional value at risk, and then builds two rolling minimum-risk portfolios. Its major finding is that the minimum-CVaR crypto portfolio outperformed the S&P 500 over the out-of-sample period, with higher Sharpe, M2, and Rachev ratios, while the minimum-variance portfolio matched the index's risk-adjusted return; both achieved this with far larger drawdowns than the index. The same modelling machinery, with normal inverse Gaussian innovations and the Esscher transform to a risk-neutral measure, produces European call and put prices for an index on those portfolios. The authors read this as evidence that an internal hedge—diversifying across crypto assets rather than hedging with outside assets—can turn individually volatile assets into an investable product, and as a fair-pricing benchmark for crypto index options should such options ever trade.","feed_headline":"Min-CVaR crypto portfolios beat the S&P 500 in backtest","feed_subtitle":"A seven-asset minimum-risk crypto portfolio out-returned the index over 2018-2019 and yields a fair option-pricing model.","key_machinery":"Two pieces carry the argument. For portfolio construction, the central object is the multivariate ARMA(1,1)-GARCH(1,1) model with Gaussian innovations and a multivariate Student-t distribution with five degrees of freedom: it converts daily log returns of seven crypto assets into 10,000 one-step-ahead scenarios per rolling window, and those scenarios feed mean-variance and CVaR optimizations that select the minimum-risk weights. For option pricing, the carrying mechanism is the Esscher transform applied to an ARMA(1,1)-GARCH(1,1) model whose innovations are normal inverse Gaussian (NIG): at each time step a one-dimensional equation tilts the physical return distribution to the unique equivalent martingale measure, and Monte Carlo averaging of discounted payoffs gives European option prices. The NIG distribution, with tail, skew, scale, and location parameters, is what lets the risk-neutral prices reproduce the short-maturity smile.","core_discovery":"The paper reports two findings. First, an equally weighted basket of the seven largest crypto assets is not the best use of the universe: over 04/04/2018 to 07/02/2019, the rolling minimum-CVaR portfolio beat the SPDR S&P 500 ETF in cumulative return during the second quarter of 2019 and recorded the best Sharpe, M2, and Rachev ratios of the three competitors, while the minimum-variance portfolio behaved similarly to SPY on those ratios; risk budgeting attributes the effect to bitcoin as the risk diversifier and EOS as the risk contributor. Second, the innovations of both minimum-risk portfolio returns fit a normal inverse Gaussian distribution inside an ARMA(1,1)-GARCH(1,1) model, and applying the Esscher transform yields unique equivalent martingale measure prices for European calls and puts on the portfolios; Monte Carlo simulation of 10,000 paths gives option smiles whose curvature increases as maturity shortens. Because no crypto index options trade, the authors cannot compare the prices with market quotes, but they claim their theoretical prices are what such options should follow once frictions and design details are adjusted for.","pith_inferences":["A stronger test would freeze the model after the in-sample window and evaluate the same weights on later data, including the 2020-2021 crypto cycle; the paper's out-of-sample period is the same window used for model selection.","Because the min-CVaR weights fluctuate sharply and crypto bid-ask spreads are wide, charging realistic transaction costs to the rebalancing rule could erase the reported edge over SPY.","If a crypto index option ever lists, the model makes a falsifiable prediction: implied volatility surfaces should look like the NIG-GARCH smiles, not flat Black-Scholes surfaces.","The risk-contribution ordering suggests a testable tilt strategy: reduce exposure to high contributors such as EOS and increase exposure to low contributors such as bitcoin, then compare the tilted basket with the optimized portfolios."],"forward_implications":["If the reported outperformance is not an artifact of the sample, diversified minimum-risk crypto portfolios become a credible equity-beating asset class built only from crypto assets.","The risk-budgeting ordering (bitcoin contributes least risk, EOS most) gives portfolio managers a concrete starting point for sizing crypto positions.","If crypto index options are introduced, the NIG-Esscher model predicts their implied volatility smiles and provides a fair-value benchmark for calls and puts.","The absence of a traded options market means the pricing claim can only be checked indirectly, for example by simulating the dynamic hedging strategy on the minimum-risk portfolios."],"supporting_citations":[{"why":"provides the mean-variance optimization from which the min-variance portfolio is taken.","marker":"Markowitz (1952)"},{"why":"defines the CVaR optimization objective used to build the min-CVaR portfolio.","marker":"Krokhmal (2002)"},{"why":"supplies the Esscher transform used to move the physical dynamics to the risk-neutral measure.","marker":"Gerber and Shiu (1994)"},{"why":"gives the GARCH-type option pricing methodology with generalized hyperbolic innovations and Monte Carlo averaging.","marker":"Chorro (2012)"},{"why":"introduces the normal inverse Gaussian distribution that fits the portfolio-return innovations.","marker":"Barndorff-Nielsen (1977)"},{"why":"provides the GARCH(1,1) conditional volatility dynamics used throughout.","marker":"Bollerslev (1986)"},{"why":"establishes that volatility and CVaR are coherent risk measures, justifying their use in risk budgeting.","marker":"Artzner et al (1999)"}],"fun_headline_variants":["Min-CVaR crypto portfolio beats S&P 500 in backtest","Crypto min-risk beats S&P 500, yields option model","Optimal crypto portfolios beat S&P 500; options priced","Min-variance crypto basket outperforms S&P 500"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the 04/04/2018 to 07/02/2019 backtest, run with zero transaction costs and with the model chosen on the same window used for evaluation, reflects how the portfolios would actually perform; if real crypto trading costs or the look-ahead in model selection are large enough, the reported edge over the S&P 500 could disappear.","fun_headline_variants_meta":{"raw":{"variants":["Min-CVaR crypto portfolio beats S&P 500 in backtest","Crypto min-risk beats S&P 500, yields option model","Optimal crypto portfolios beat S&P 500; options priced","Min-variance crypto basket outperforms S&P 500"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000827,"raw_usage":{"total_tokens":3690,"prompt_tokens":1094,"completion_tokens":2596,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":710,"completion_tokens_details":{"reasoning_tokens":2523}},"tokens_in":710,"tokens_out":2596,"duration_ms":18722,"temperature":1.0,"reasoning_tokens":2523,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:14:47.608377+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the same rolling optimization on data after July 2019 (or between 2017 and 2018) and record the min-CVaR portfolio's cumulative return net of realistic crypto bid-ask spreads and rebalancing fees; if it no longer exceeds the S&P 500 on Sharpe or cumulative return, the paper's major finding is specific to the original window and frictionless assumption.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"provides the mean-variance optimization from which the min-variance portfolio is taken."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"defines the CVaR optimization objective used to build the min-CVaR portfolio."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the Esscher transform used to move the physical dynamics to the risk-neutral measure."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"gives the GARCH-type option pricing methodology with generalized hyperbolic innovations and Monte Carlo averaging."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"provides the GARCH(1,1) conditional volatility dynamics used throughout."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"establishes that volatility and CVaR are coherent risk measures, justifying their use in risk budgeting."}],"review_version":1}