{"id":"6ae91ca0-b691-4976-9bb8-805c3ce34bff","arxiv_id":"2412.02654","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":7,"one_line_summary":"A 90/10 traditional/crypto portfolio, diluted with cash to a target risk level, matched a full risk-allocation optimizer and beat an equities-only portfolio in a 2017-2024 backtest.","lead":"This paper tests whether simple portfolio rules can safely add crypto assets to a traditional stock portfolio, and finds a fixed 90/10 split with cash adjustments works as well as a more complex optimizer. The result suggests investors can get crypto diversification benefits without machine-learning models.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claimed Sharpe improvement of DD90/10 (1.06 vs 0.73) may not survive realistic transaction costs because the strategy rebalances daily and the backtest omits all trading costs.","rationale":"The reader's weakest_assumption bundles two issues: in-sample selection of the 90/10 rule and the absence of transaction costs. I agree with both, but I identify the transaction-cost omission as the single most load-bearing concern because it directly undermines the practical, after-cost performance claim that is central to the paper. The in-sample selection is a real concern, but the 90/10 split is partly dictated by the 10% maximum crypto constraint and is a simple, interpretable rule; moreover, the paper explicitly claims the results are not sensitive to the risk limit. The cost omission, by contrast, is an unmodeled mechanism that could quantitatively reverse the headline Sharpe ordering, and it is trivially testable from the publicly available code. My recommendation remains CONDITIONAL, matching the reader's verdict, because the paper's methodological contribution (CRA via convex optimization) is sound and the empirical shortfall can be addressed with additional analysis. If a cost-aware backtest shows the Sharpe advantage evaporates, the verdict should move toward REJECT; if it survives, the conditional accept should stand. The reader's verdict does not need adjustment; my concern reinforces the need for the conditional requirement of cost-aware validation.","tokens_in":11167,"tokens_out":6552,"duration_ms":69350,"concrete_test":"Reproduce the DD90/10 and industry-only backtests from the provided GitHub repository. For each trading day, compute the one-way turnover as the sum over assets of |w_t^pre − w_t^post| (with cash included), then annualize. Apply a grid of one-way transaction cost parameters (e.g., 0, 5, 10, 25, 50, 100 bps) to both strategies, using the same cost for all assets initially and a higher cost for crypto as a robustness case. If the difference in Sharpe between DD90/10 and the industry portfolio falls below 0.1 at any one-way cost ≤ 50 bps, the claim that the simple rule robustly outperforms is not supported. Report both the break-even cost at which the Sharpe difference is zero and turnover statistics.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's headline empirical claim is that DD90/10 delivers Sharpe 1.06 versus 0.73 for the industry-only portfolio (Table 5) and that adding a 10% crypto weight increases return and Sharpe 'significantly' without significantly increasing volatility or drawdown (Conclusion). This is an after-cost performance claim about a practical rule. Yet the entire backtest, for both CRA and DD90/10, is executed with daily rebalancing (§4.1, §5.1) and no transaction costs, spreads, or slippage are modeled anywhere in the paper. Daily rebalancing of a 90/10 risky sleeve to fixed relative weights, combined with day-to-day changes in the EWMA/GARCH volatility estimate that drive the cash dilution, implies a high turnover: substantial fractions of the portfolio will be traded each day whenever asset returns diverge. A conservative round-trip cost of 20–50 basis points on traded notional would plausibly reduce annual return by over 1–2 percentage points, a material share of the reported 10.4% return. The Sharpe ratio could drop from 1.06 toward the industry-only 0.73, reversing the paper's central practical conclusion. The omission also asymmetrically affects the comparison, since the industry-only portfolio also trades daily but with less crypto exposure (crypto is the most volatile component, generating more weight drift and turnover). Without a cost model, the 'simple and effective' claim is not established.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper considers portfolio construction that combines four traditional industry portfolios with two crypto assets (BTC, ETH). It proposes a constrained risk allocation (CRA) method that minimizes cash subject to exact risk-contribution targets, a total risk limit, and weight constraints, and shows that the nonconvex risk-allocation constraints reduce to a ray parameterized by one scalar, so the CRA problem is solved by a convex program followed by a closed-form scaling. The empirical section backtests CRA and a fixed 90/10 traditional/crypto sleeve with dynamic cash dilution (DD90/10) over September 2017 to September 2024, reporting that DD90/10 achieves annualized return 10.4%, volatility 9.8%, Sharpe 1.06, and drawdown 19.9%, versus 6.0%, 8.2%, 0.73, and 12.5% for the industry-only portfolio. The paper concludes that simple risk allocation suffices to integrate crypto assets and that a modest 10% crypto weight improves risk-adjusted performance without materially increasing risk.","tokens_in":11497,"tokens_out":3896,"duration_ms":42844,"significance":"If the empirical claims survive scrutiny, the paper offers a genuinely simple, implementable portfolio rule and supports the position that crypto assets do not require machine-learning-based methods. The convex optimization reduction in Section 3 is clean and correctly derived under the stated assumptions, and the paper provides a data/code repository, which are concrete strengths. The main significance is tempered by the fact that the central empirical claim rests on a single in-sample backtest with no transaction costs and with the DD90/10 rule selected from the same sample on which it is evaluated. The optimization contribution itself is solid and reproducible, but the practical 'simple and effective' conclusion needs additional empirical support.","major_comments":[{"comment":"The headline comparison is made before any transaction costs, although both CRA and DD90/10 are rebalanced daily and the cash-dilution weight is revised whenever the volatility estimate changes. No turnover is reported. Because crypto is the most volatile component, its weight drifts quickly, so the 90/10 sleeve will generate substantial daily trades. A conservative round-trip cost of 20–50 basis points on traded notional could plausibly reduce annualized return by more than 1–2 percentage points, which is a material share of the reported 10.4% return and could reverse the Sharpe ordering relative to the 0.73 industry-only portfolio. Please add a transaction-cost model, report turnover, or at minimum provide after-cost results for both DD90/10 and the industry-only benchmark.","section":"§4.1, §5.1, Table 5"},{"comment":"The DD90/10 rule is reverse-engineered from the full-sample CRA holdings: Figure 9 is inspected over the entire 2017–2024 period, and the fixed 90/10 split is then evaluated on that same period in Table 5. This makes the comparable performance of DD90/10 and CRA partly an artifact of in-sample model selection. The robustness claim would be considerably strengthened by an out-of-sample evaluation, a walk-forward procedure, or at least a subperiod analysis in which the 90/10 allocation is chosen before the evaluation window.","section":"§5, Figure 9, Table 5"},{"comment":"The conclusion states that adding a 10% crypto weight increases return and Sharpe ratio 'significantly,' but no confidence intervals, bootstrap resampling, or other uncertainty quantification is provided. With a single seven-year price path, the reported Sharpe difference (1.06 versus 0.73) may be within sampling variability, especially given the heavy tails documented in Section 1. Please add block-bootstrap intervals or a per-year/subperiod analysis to support the word 'significantly.'","section":"§4.3, §6, Table 5"},{"comment":"The paper asserts parameter insensitivity and robustness to alternative volatility estimators without reporting the supporting evidence. In Section 4.1, the risk limit, maximum crypto weight, and EWMA half-lives are said to be 'chosen as reasonable values' with results 'not sensitive to these choices,' and in Section 5.1, several other volatility estimators are said to give similar results. Since these are free parameters and the DD90/10 performance is the central claim, a sensitivity table or figure is needed.","section":"§4.1, §5.1"}],"minor_comments":[{"comment":"The word 'portolios' should be 'portfolios.'","section":"§4.3"},{"comment":"The word 'contrstruction' should be 'construction.'","section":"§6"},{"comment":"In the drawdown definition, 'form' should be 'from': 'the maximum fractional drop in value from a previous high.'","section":"§4.2"},{"comment":"The word 'attribtutions' should be 'attributions.'","section":"§4.4"},{"comment":"Please clarify how the 2565 calendar days and 1729 trading days are used in the backtest, and how weekend/holiday crypto gains are incorporated when rebalancing only on trading days.","section":"§4.1"},{"comment":"The characterization that weights satisfying the risk-allocation constraints form a ray generated by the convex program (3) is cited to [BV24]; since this is the key mathematical step, a short proof or an explicit statement of the required positive-definiteness assumption would make the note more self-contained.","section":"§3.2"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is clearly written and provides code and data, which is commendable. The Section 3 derivation is correct but is not the main source of uncertainty; the empirical claim in Table 5 is. The absence of transaction costs and the in-sample selection of DD90/10 are the two issues that must be addressed before the paper can support its practical conclusion. I saw no indication of misrepresentation, but the in-sample selection should be disclosed more prominently. With a credible cost analysis and out-of-sample check, the paper could be suitable for a practitioner-oriented finance outlet."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the paper is a clean, honest note. The math is standard risk-parity scaling, correctly derived and properly cited to [BV24] and [FP15] — no new theory there. What is genuinely new is the empirical observation that a fixed 90/10 traditional/crypto sleeve, diluted with cash to a target risk, tracks the full constrained risk allocation (CRA) optimizer on 2017-2024 daily data. That is a useful simplification for practitioners. The authors also ship code and data, and the Shapley attribution section is a nice addition. They do not oversell; the conclusions are modest.\n\nThe soft spots are real but not fatal. First, the DD90/10 rule is selected after inspecting the CRA weights on the full sample (Figure 9) and then evaluated on the same sample (Table 5). That is in-sample model selection. It is not definitionally circular — the CRA solution comes from stated assumptions — but the performance comparison is partly reverse-engineered. A walk-forward or a split-sample exercise would fix it. Second, the backtest is daily rebalanced with no transaction costs, spreads, or slippage. That matters more than the paper acknowledges: crypto is the most volatile sleeve, so it generates the most weight drift, and daily cash dilution adds turnover. A 20-50bp round-trip cost could plausibly shave 1-2pp off the reported 10.4% return and push the Sharpe from 1.06 toward the industry-only 0.73. The stress-test note lands. Third, the sensitivity claims — 'results are not sensitive', 'all gave similar results' — are asserted but not shown. For a note that is tolerable; for a peer-reviewed claim they should be backed by a table or a sentence on the range tried. Also, the conclusion says the 10% crypto weight does not 'significantly' increase drawdown, but the reported drawdown goes from 12.5% (industries) to 19.6% (combined). That is a large relative jump and the wording should be fixed.\n\nBottom line: this is a useful practitioner result, not a breakthrough. It deserves a serious referee. I would send it out, and ask the authors for an out-of-sample or walk-forward validation, a transaction-cost model, the sensitivity evidence, and a corrected drawdown statement. If the Sharpe gap survives those, it is a solid applied-finance note. I'd cite it in the portfolio-construction context.","headline":"Clean, honest practitioner note with standard math and a useful empirical simplification; the headline Sharpe gap is in-sample-selected and cost-blind, so it needs a walk-forward and a transaction-cost model before I'd trust it.","tokens_in":12017,"tokens_out":3598,"would_cite":true,"duration_ms":34141,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["91G10","90C25"],"pacs":[],"model":"deepseek-v4-flash","headline":"A fixed 90/10 traditional-and-crypto allocation, diluted with cash to a target risk, matches a full risk-allocation optimizer in backtests, with Sharpe rising from 0.73 to 1.06.","keywords":["portfolio construction","crypto assets","risk parity","constrained risk allocation","dynamic cash dilution","convex optimization","Sharpe ratio"],"falsifier":"Run a genuinely out-of-sample test: fix the 90/10 split and the 10% risk target using data only up to, say, January 2020, then backtest the rule on the remaining period with realistic transaction costs on daily rebalancing; if its Sharpe ratio no longer beats the 0.73 of the industry-only portfolio, the central claim fails.","tokens_in":10965,"feed_emoji":"📈","tokens_out":9313,"duration_ms":85863,"temperature":0.7,"pith_summary":"The paper argues that crypto assets can be added to a conventional portfolio with ordinary risk-allocation tools, despite their extreme volatility and heavy tails. Its central demonstration is a constrained risk-allocation (CRA) portfolio that sets risk contributions and scales exposure to a 10% annualized risk cap; adding 10% crypto raises the backtested Sharpe ratio from 0.73 for an industry-only portfolio to 1.00, with similar volatility and drawdown. Inspecting those weights suggests an even simpler rule, DD90/10: hold 90% traditional industry portfolios and 10% crypto in fixed relative weights, and dilute with cash each day to hit the same risk target. In the same backtest this rule reports Sharpe 1.06 with an EWMA volatility estimate and 1.04 with GARCH, slightly better than the optimizer. The practical point is that investors may not need return forecasts or machine-learning models to benefit from crypto exposure.","feed_headline":"Simple 90/10 crypto rule lifts Sharpe from 0.73 to 1.06","feed_subtitle":"A fixed 90% stocks, 10% crypto mix, scaled by cash to a 10% risk target, matches a full optimizer in backtests.","key_machinery":"The central object is the constrained risk-allocation (CRA) problem: choose weights $w \\ge 0$ and cash $c$ to minimize cash holdings subject to risk contributions $w_i(\\Sigma w)_i = \\rho_i w^T\\Sigma w$, a risk cap $w^T\\Sigma w \\le \\sigma^2$, and weight constraints. Its solution rests on the fact that the risk-allocation equations force $w = \\alpha x^\\star$, where $x^\\star$ is the unique minimizer of $(1/2)x^T\\Sigma x - \\sum_i \\rho_i \\log x_i$, so the full problem collapses to a scalar scaling rule, $\\alpha^\\star = \\min\\{1/(\\mathbf{1}^T x^\\star),\\, \\sigma/\\sqrt{(x^\\star)^T\\Sigma x^\\star},\\, g_i/(F x^\\star)_i\\}$. The takeaway rule DD90/10 replaces the optimizer with fixed 90/10 relative weights and uses only a volatility estimate — a 10-day half-life EWMA or a GARCH(1,1) model — to set the cash dilution each day. The mechanism that carries the argument is this dynamic cash dilution, which converts a fixed risky allocation into one that respects a target risk level.","core_discovery":"The paper's discovery is that a portfolio built by constrained risk allocation — specifying the fraction of total risk each asset should contribute, then scaling the whole position to a risk cap — handles crypto assets without any special machinery. The weights are found by solving one convex optimization problem, and the risk-allocation constraints reduce the search to a single scaling factor. Looking at the resulting relative weights over the sample, the authors observe that the optimizer itself keeps roughly 10% in crypto and 90% in the four industry portfolios, which motivates DD90/10, a fixed-relative-weight portfolio whose only time-varying decision is how much cash to hold. In the 2017–2024 backtest, this simple rule slightly outperforms the CRA optimizer on Sharpe ratio while keeping volatility and drawdown at the target levels, and both crypto-inclusive portfolios beat the industry-only portfolio. The conclusion is that the extreme statistical properties of crypto returns do not require abandoning standard risk-based construction.","pith_inferences":["The 90/10 split was selected by looking at the full-sample CRA weights and then evaluated on that same sample; an out-of-sample test that fixes the split before the evaluation period would tell whether the simplicity is driving the result or merely being rewarded in-sample.","Daily rebalancing is assumed to be free; including transaction costs would erode the Sharpe advantage, particularly for crypto assets with wide spreads and in volatile periods that force large cash adjustments.","The specific 90/10 split and the four chosen industry portfolios are calibrated to this six-asset universe; other traditional assets or crypto tokens could yield a different best split, so the rule's sharpness as a universal prescription is untested.","If the mechanism is really the cash-dilution risk control rather than the exact 90/10 split, the same template should transfer to other volatile satellite assets, which is a testable extension."],"forward_implications":["A 10% crypto allocation in a risk-allocation portfolio raised the backtested Sharpe ratio from 0.73 to 1.00 without meaningfully increasing volatility or drawdown.","The DD90/10 rule needs no expected-return forecasts: its only inputs are a fixed relative weight and a daily volatility estimate.","Because the CRA problem reduces to one convex optimization plus a scalar scaling, the full method can be rebalanced every trading day at low computational cost.","The results are reported under a 10% annualized risk cap, a 10% cap on combined crypto weight, and daily rebalancing over September 2017 to September 2024.","The paper's comparison suggests that a fixed 90/10 split with cash dilution performs at least as well as the optimizer it was extracted from."],"supporting_citations":[{"why":"Supplies the risk-parity idea that the constrained risk-allocation problem extends.","marker":"[Qia11]"},{"why":"Provides the convex optimization derivation showing that risk-allocation weights form a ray through the minimizer of a log-barrier problem.","marker":"[BV24]"},{"why":"Supplies the iterated EWMA covariance estimation used for the risk model and the volatility estimates.","marker":"[JOP+23]"},{"why":"Provides the four industry portfolio return series used in the backtest.","marker":"[Fre24]"},{"why":"Defines the mean-variance portfolio problem that the paper deliberately bypasses by using risk allocation without expected-return estimates.","marker":"[Mar52]"}],"fun_headline_variants":["90/10 crypto allocation rivals full optimizer, boosts Sharpe","Simple 90/10 rule beats optimizer in crypto portfolio test","Match optimizer with 90/10 crypto, only cash scaling needed","Crypto-heavy 90/10 portfolio matches optimizer with cash scaling","90/10 rule lifts Sharpe from 0.73 to 1.06 in crypto backtest"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The fixed 90/10 split is chosen by inspecting the optimized weights over the full backtest period and then judged on that same period, so the headline Sharpe gain assumes this way of choosing the rule is not flattering the result and that ignoring trading costs would not erase the gain.","fun_headline_variants_meta":{"raw":{"variants":["90/10 crypto allocation rivals full optimizer, boosts Sharpe","Simple 90/10 rule beats optimizer in crypto portfolio test","Match optimizer with 90/10 crypto, only cash scaling needed","Crypto-heavy 90/10 portfolio matches optimizer with cash scaling","90/10 rule lifts Sharpe from 0.73 to 1.06 in crypto backtest"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000607,"raw_usage":{"total_tokens":2772,"prompt_tokens":829,"completion_tokens":1943,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":445,"completion_tokens_details":{"reasoning_tokens":1848}},"tokens_in":445,"tokens_out":1943,"duration_ms":14156,"temperature":1.0,"reasoning_tokens":1848,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T23:12:10.127028+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a genuinely out-of-sample test: fix the 90/10 split and the 10% risk target using data only up to, say, January 2020, then backtest the rule on the remaining period with realistic transaction costs on daily rebalancing; if its Sharpe ratio no longer beats the 0.73 of the industry-only portfolio, the central claim fails.","supporting_citations":[],"review_version":1}