{"id":"96a19515-00a3-4bd6-90ec-063db197c7cb","arxiv_id":"2507.23414","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"Applying RCMSE and MF-DFA to daily log-returns, the authors find Bitcoin has the highest multiscale entropy sum and the widest multifractal spectrum among the four assets.","lead":"This paper measures how unpredictable Bitcoin, GBP/USD, gold, and natural gas are using two standard complexity tools, entropy and multifractal fluctuation analysis. It reports that Bitcoin looks most complex overall, mostly because shuffling its returns changes the statistics more than it does for other assets.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The ranking of Bitcoin as most complex is not established because the RCMSE complexity score integrates up to scale 100, where coarse-grained series are only ~37 points long, below the paper's own 10^m–30^m guideline for m=3, and Bitcoin's advantage appears only at these large scales.","rationale":"The paper is a conventional application of two established estimators; I credit it for giving explicit parameter choices and for using RCMSE rather than plain MSE. The descriptive numbers are internally consistent with the figures. My concern is not that the computations are wrong but that the specific aggregate used to rank the assets is dominated by a regime where the estimator is known to be unreliable. The reader's conditional verdict already asks for uncertainty quantification and matched windows; my check narrows this to the τ dependence of the RCMSE sum. If the scale-restricted recomputation keeps Bitcoin highest, I would regard the claim as supported for small scales, and the conditional verdict could move toward acceptance; if not, the central claim should be dropped or reframed. The sample-window mismatch flagged by the reader is a real secondary confound: Bitcoin's sample begins in 2016 whereas the others begin in 2010, so the comparison conflates asset identity with market regime. But even a matched-window rerun would still need to address the τ = 100 entropy estimates. The issue is methodological and testable; I do not see grounds for rejection, only for strengthening the conditions already imposed by the reader.","tokens_in":11535,"tokens_out":7172,"duration_ms":85502,"concrete_test":"Recompute the RCMSE complexity in Table 2 with τ_max = 3 (the largest scale satisfying the paper's own 10^m length criterion) and also with τ_max = 30 and τ_max = 50 as sensitivity checks, keeping m = 3 and r = 0.15σ; report block-bootstrap 95% intervals for the summed complexity for every asset. If Bitcoin is not highest for τ_max = 3, or if its margin is inside the bootstrap interval, the claim that Bitcoin is the most complex is unsupported and the manuscript should restrict conclusions to scales where RCMSE estimates are reliable.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The headline ranking in Table 2 is not a robust finding because the RCMSE \"complexity\" is the sum of entropy over scales τ = 1..100 and Bitcoin's lead appears only at the upper end of this range (Fig. 6). At those scales the coarse-grained series are very short: for N = 3,730, τ = 100 leaves 37 points, and τ > 3 falls below the 10^m–30^m series-length rule quoted in §2.1 for m = 3 (1,000–27,000 points). RCMSE reduces the incidence of undefined entropies, but it does not remove finite-sample bias and variance of sample-entropy counts at lengths of 37. Bitcoin's total (74.66) exceeds gold (67.88) by about 10%, a margin easily within the expected uncertainty of such estimates; no bootstrap intervals, surrogate tests, or parameter sensitivity are reported. Since Bitcoin has the lowest entropy at small scales, its entire lead is generated by scales where the estimator is least reliable. Matching sample periods would remove a separate confound, but it would not fix this estimator-reliability problem. The MF-DFA width is consistent with the claim but carries its own finite-sample uncertainty, and the shuffled-series analysis shows reduced widths for all assets, not only Bitcoin.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript applies Refined Composite Multiscale Sample Entropy (RCMSE) and Multifractal Detrended Fluctuation Analysis (MF-DFA) to daily log-return series of Bitcoin, GBP/USD, gold, and natural gas, each containing 3,730 points, and reports two complexity rankings. The RCMSE-based complexity, defined as the sum of entropy values over 100 coarse-graining scales, places Bitcoin highest (74.66) followed by gold (67.88), GBP/USD (67.24), and natural gas (51.48). The MF-DFA singularity-spectrum width also ranks Bitcoin highest (0.62) versus 0.50, 0.44, and 0.21 for the others. The authors interpret these results as evidence that Bitcoin displays the greatest complexity and that this complexity reflects nonlinear correlations in its log returns.","tokens_in":11799,"tokens_out":5125,"duration_ms":57688,"significance":"If the ranking were established with proper uncertainty quantification and comparable sampling windows, the finding would be a useful comparative stylized fact for cryptocurrency versus traditional financial markets, and the combination of RCMSE and MF-DFA is a reasonable and standard toolkit for the question. The paper clearly explains the algorithms and presents the results in an accessible way. However, the headline comparisons currently rest on entropy estimates at scales where the coarse-grained series are far shorter than the method's own length guideline, on unmatched historical windows, and on point estimates without confidence intervals. The manuscript also provides no code or data, so the central quantitative claims are not independently verifiable as presented.","major_comments":[{"comment":"The complexity score is the sum of RCMSE values over scales τ = 1 to 100. Section 2.1 states that meaningful sample entropy requires series lengths between 10^m and 30^m; for m = 3 this means 1,000 to 27,000 points. With N = 3,730, only τ ≤ 3 satisfies the lower bound. Scales 4 through 100 use coarse-grained series of length 932 down to 37 points, for which the estimates are subject to strong finite-sample bias and variance. RCMSE reduces the incidence of undefined entropies, but it does not eliminate this problem. The paper's own Fig. 6 shows that Bitcoin has the lowest entropy at small scales and the highest only at large scales, so its entire lead is generated by scales where the estimator is least reliable. The authors should report complexity computed over scales satisfying the stated guideline, or use an embedding dimension for which the guideline is met over a defensible scale range, and should accompany any summed score with bootstrap confidence intervals.","section":"Section 4.2, Eq. (6), Fig. 6, Table 2"},{"comment":"The four series are not directly comparable as collected. Bitcoin begins in 2016, while GBP/USD, gold, and natural gas begin in 2010, and all series are truncated to 3,730 points. The Bitcoin sample therefore covers a different market regime, including the 2017 and 2021 cryptocurrency boom-bust episodes, whereas the other assets' samples include the 2010s sovereign-debt and commodity cycles. Any of the reported differences could be an artifact of this window mismatch. The authors should show that the ranking is robust to using a common sample period, for example 2016 onward for all assets, or at least provide a matched-window sensitivity analysis.","section":"Section 3, Fig. 3"},{"comment":"No uncertainty quantification is given for the central point estimates. Bitcoin's RCMSE complexity (74.66) exceeds gold's (67.88) by about 10%, and its singularity-spectrum width (0.62) exceeds GBP/USD's (0.50) by about 24%, but without bootstrap intervals, surrogate-data tests, or parameter sensitivity analysis these differences cannot be distinguished from estimation error. This is especially important because Table 4 reports Hurst exponents with errors while Tables 2 and 3 do not. The authors should provide confidence intervals for the summed RCMSE values and the spectrum widths, and ideally a statistical test of the ranking.","section":"Tables 2 and 3"},{"comment":"The shuffled-data analysis does not support the claim that nonlinear correlations are distinctive to Bitcoin. Figure 9 shows that the singularity-spectrum width decreases after shuffling for all four assets, not only for Bitcoin, and the text acknowledges a 'notable reduction' across all assets. The claim that Bitcoin alone shows a significant entropy increase after shuffling is based on a bar chart with no error bars or statistical test, despite the statement that each series was shuffled 100 times. The authors should report the mean and standard deviation of the shuffled entropies and spectrum widths per asset, and test whether Bitcoin's shuffle-induced change is significantly larger than those of the other assets.","section":"Section 4.2, Fig. 7 and Section 4.3, Fig. 9"},{"comment":"The MF-DFA results are not reproducible because the parameter values are not reported. The manuscript does not state the range of q values used, the order of the detrending polynomial, or the range of segment lengths s over which the fluctuation functions were fitted. These choices materially affect the generalized Hurst exponents and the singularity-spectrum width, so without them Table 3 and Fig. 8 cannot be checked. The authors should list the exact parameters and ideally provide the code or a pseudocode description.","section":"Section 4.3, Eqs. (7)-(14)"}],"minor_comments":[{"comment":"The asterisks in the table captions are unexplained; either define them or remove them.","section":"Table 2 and Table 3 captions"},{"comment":"The table header promises 'complexity values across 100 time scales, as well as the entropy at the first scale,' but the first-scale entropy is not shown in the table; either add these values or revise the heading.","section":"Section 4.2, Table 2"},{"comment":"The 'complexity' used in Table 2 is defined only in prose as a sum over scales; the authors should define it explicitly, for example as C = Σ_{τ=1}^{100} RCMSE(x, τ, m, r), to avoid ambiguity about whether scale 1 is included.","section":"Section 4.2"},{"comment":"The Hurst exponent errors are given inconsistently: the text reports H = 0.51 ± 0.01 for Bitcoin, while the table reports 0.51 ± 0.1, and similar discrepancies appear for the other assets. These should be reconciled.","section":"Table 4 and Fig. 11"},{"comment":"The figure shows entropy before and after shuffling but provides no error bars, even though the text states that shuffling was repeated 100 times; plotting means with standard deviations would make the claimed difference visible.","section":"Fig. 7"},{"comment":"The statement that Hurst exponents near 0.5 indicate 'a lack of linear autocorrelation' is too strong; a Hurst exponent near 0.5 indicates an absence of long-range dependence, and short-range linear autocorrelation can still be present. The wording should be softened.","section":"Section 4.4"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the scope of q-fin.ST and addresses a question of interest, but the headline ranking is currently supported mainly by entropy estimates in a parameter regime that the paper itself identifies as unreliable, and by unmatched sampling windows. I recommend major revision rather than rejection because the weaknesses are addressable with additional robustness analyses, matched-period comparisons, and uncertainty quantification. I would also encourage the editor to ask for the data and code to be made available, as the quantitative claims are otherwise hard to verify."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a textbook application of two standard complexity tools (RCMSE, MF-DFA) to four log-return series, and the headline result—Bitcoin is the most complex—is plausible but not actually established by the evidence as reported. The paper's real value is as a clean comparative snapshot; the main weakness is that the RCMSE complexity score is a sum over 100 scales, and the ranking is decided at scales where the estimator is operating outside its validity range.\n\nThe authors do some things well. They describe the algorithms accurately, use standard parameter choices (m=3, r=0.15*SD), and are transparent that 'complexity' is defined operationally as the sum of entropy over scales. The shuffled-series checks are a reasonable first pass, and the MF-DFA result—Bitcoin's wider singularity spectrum (Δα = 0.62 vs 0.21–0.50)—is consistent with existing literature on multifractality in crypto assets. So the directional claim is not a crank result.\n\nThe soft spots are significant, but proportionate. Tables 2 and 3 have no error bars, bootstrap intervals, or surrogate significance tests; the RCMSE margin (74.66 vs 67.88) is small and could easily be within finite-sample variability. More importantly, the RCMSE complexity integrates up to scale 100, where coarse-grained series are only 37 points long, far below the paper's own 10^m guideline (1,000 points for m=3). Bitcoin's lead appears only at those unreliable large scales; at small scales it actually has the lowest entropy. That makes the complexity ranking an artifact of the aggregation range. The window mismatch (Bitcoin starts 2016, others 2010) is a secondary confound, and the absence of code/data prevents independent checking. The MF-DFA half is more robust, but it too lacks uncertainty quantification, and the shuffle test reduces spectral width for all assets, not just Bitcoin, so the specificity of the nonlinear-correlation claim is weak.\n\nThis paper is for readers who want a quick, standard-method comparison of these four assets, especially the multifractal spectrum widths. It is not a methodological or conceptual advance. My recommendation: it is a marginal empirical note, but the flaws are fixable. A serious referee could push for matched sample periods, bootstrap or surrogate confidence intervals, and a scale cutoff consistent with the paper's own length guideline. As it stands, the headline ranking is not robust enough to stand on its own.","headline":"Plausible directional claim, but the RCMSE complexity ranking is fragile because it leans on scales where the estimator is invalid; the MF-DFA finding is on firmer ground.","tokens_in":12356,"tokens_out":2953,"would_cite":false,"duration_ms":30071,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Bitcoin's daily log returns are the most complex of the four assets studied, under both RCMSE and MF-DFA, with the edge attributed to nonlinear correlations.","keywords":["RCMSE","MF-DFA","multifractal spectrum","sample entropy","Bitcoin","financial time series","nonlinear correlations","Hurst exponent"],"falsifier":"Run the same RCMSE and MF-DFA computations on the four assets over the identical calendar window from 2016 to 2024. If Bitcoin's complexity score and singularity-spectrum width drop into the range of the other assets, the claimed ranking is an artifact of the mismatched histories rather than a stable property of Bitcoin log returns.","tokens_in":11288,"feed_emoji":"📊","tokens_out":5704,"duration_ms":59890,"temperature":0.7,"pith_summary":"This paper sets out to show that Bitcoin's daily log-return series is more complex than the log returns of GBP/USD, gold, and natural gas, and that the extra complexity traces to nonlinear correlations rather than to simple volatility or linear autocorrelation. It applies two independent measures: refined composite multiscale sample entropy, which sums regularity across many coarse-grained time scales, and multifractal detrended fluctuation analysis, which measures how fluctuation scaling varies across moment orders. Both rank Bitcoin first: its RCMSE complexity score is 74.66 versus 51.48 to 67.88 for the others, and its multifractal spectrum width is 0.62 versus 0.21 to 0.50. Bitcoin shows the lowest entropy at short scales and the highest at long scales, so the claim is scale-dependent rather than a blanket statement of unpredictability. The result matters because complexity assessment bears on predictability and risk for investors and on which assets are adequately described by standard efficient-market models.","feed_headline":"Bitcoin is the most complex of four assets, by two independent measures","feed_subtitle":"Both methods point to nonlinear correlation structure that standard random-walk models would miss.","key_machinery":"Two named measures carry the argument. RCMSE, refined composite multiscale sample entropy, coarse-grains the series into multiple scale-factor windows, counts matching template vectors across all coarse-grained versions, and defines complexity as the sum of entropy over scales; it is chosen because ordinary multiscale entropy becomes undefined for short series at large scales. MF-DFA, multifractal detrended fluctuation analysis, builds a cumulative profile, splits it into segments, removes local polynomial trends, and computes q-th order fluctuation functions whose scaling exponents yield the singularity spectrum; its width, delta alpha, is the fractal complexity measure. The two methods are corroborated by shuffling tests, which remove temporal structure and show that Bitcoin's entropy and spectrum change far more than the other assets' do.","core_discovery":"On the paper's own terms, the central discovery is that Bitcoin's log-return series combines short-scale regularity with long-scale irregularity to a degree the other assets do not show. At scale one its sample entropy is the lowest of the four, suggesting stronger short-horizon regularity, yet summed over 100 scales its RCMSE complexity reaches 74.66, well above 67.88 for gold, 67.24 for GBP/USD, and 51.48 for natural gas. The multifractal singularity spectrum is also widest for Bitcoin, with width 0.62 versus 0.50, 0.44, and 0.21, indicating a richer range of fluctuation strengths across time scales. Hurst exponents sit near 0.5 for all assets, so the authors interpret the entropy increase after shuffling and the spectrum narrowing after shuffling as evidence that Bitcoin's complexity comes from nonlinear autocorrelations rather than from linear memory.","pith_inferences":["Editorial inference: because Bitcoin's window starts in 2016 while the other assets start in 2010, the ranking should be re-tested on a common 2016-2024 window; the observed edge may reflect the crypto bull and bear regime rather than a stable asset-level property.","Editorial inference: the shuffling comparison could be turned into a formal significance test by generating many surrogate series and counting how often shuffled complexity exceeds the observed value, giving error bars for the Bitcoin-versus-others gap.","Editorial inference: the parameter choices of embedding dimension 3 and tolerance 0.15 times the standard deviation may affect the RCMSE ranking; a sensitivity sweep over embedding dimensions and tolerances would tell whether Bitcoin's lead is robust."],"forward_implications":["If Bitcoin's complexity ranking is right, short-horizon forecasts may be comparatively reliable for Bitcoin while long-horizon behavior is comparatively hard to predict.","Risk models for Bitcoin that assume simple random-walk or linear-autocorrelation dynamics would miss the nonlinear dependencies the shuffling tests expose.","The near-0.5 Hurst exponents across all assets mean standard detrended fluctuation analysis alone would not separate Bitcoin from the others; the multifractal and multiscale-entropy layers are what carry the ranking.","For GBP/USD, gold, and natural gas, the small changes after shuffling suggest their log-return series may be adequately described by random or linear processes, so complexity-based risk adjustments may matter less there."],"supporting_citations":[{"why":"Supplies the MF-DFA method that yields the generalized Hurst exponents and singularity spectra used for the fractal complexity ranking.","marker":"[21]"},{"why":"Introduces RCMSE and its coarse-graining scheme, the method that produces the summed multiscale entropy complexity scores.","marker":"[25]"},{"why":"Provides the original multiscale entropy coarse-graining idea that RCMSE refines for short time series.","marker":"[7]"},{"why":"Defines sample entropy, the matching-vector regularity measure at the core of RCMSE.","marker":"[4]"},{"why":"Introduces detrended fluctuation analysis, the monofractal precursor that MF-DFA extends.","marker":"[20]"},{"why":"Documents the undefined-entropy problem of ordinary multiscale entropy at large scales, motivating the choice of RCMSE.","marker":"[24]"},{"why":"Provides a prior multifractal comparison across cryptocurrencies and forex markets that this study extends to the four selected assets.","marker":"[23]"}],"fun_headline_variants":["Bitcoin tops four-asset complexity test using two metrics","Dual analysis ranks Bitcoin as the most complex market","Bitcoin's multifractal and entropy signatures beat gold and FX","Bitcoin shows richest complexity across scales in four markets","Two independent measures crown Bitcoin the most complex asset"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing assumption is that the four series are directly comparable even though Bitcoin's data begin in 2016 while the others begin in 2010 and all are truncated to 3,730 points; if Bitcoin's higher complexity is a property of that later period rather than of Bitcoin itself, the ranking collapses.","fun_headline_variants_meta":{"raw":{"variants":["Bitcoin tops four-asset complexity test using two metrics","Dual analysis ranks Bitcoin as the most complex market","Bitcoin's multifractal and entropy signatures beat gold and FX","Bitcoin shows richest complexity across scales in four markets","Two independent measures crown Bitcoin the most complex asset"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000312,"raw_usage":{"total_tokens":1724,"prompt_tokens":846,"completion_tokens":878,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":462,"completion_tokens_details":{"reasoning_tokens":799}},"tokens_in":462,"tokens_out":878,"duration_ms":9698,"temperature":1.0,"reasoning_tokens":799,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T10:46:29.607863+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same RCMSE and MF-DFA computations on the four assets over the identical calendar window from 2016 to 2024. If Bitcoin's complexity score and singularity-spectrum width drop into the range of the other assets, the claimed ranking is an artifact of the mismatched histories rather than a stable property of Bitcoin log returns.","supporting_citations":[{"cited_title":"Physica A 316(1-4), 87–114 (2002)","cited_arxiv_id":null,"evidence_quote":"Supplies the MF-DFA method that yields the generalized Hurst exponents and singularity spectra used for the fractal complexity ranking."},{"cited_title":"Physics Lett ers A 378(20), 1369–1374 (2014)","cited_arxiv_id":null,"evidence_quote":"Introduces RCMSE and its coarse-graining scheme, the method that produces the summed multiscale entropy complexity scores."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the original multiscale entropy coarse-graining idea that RCMSE refines for short time series."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines sample entropy, the matching-vector regularity measure at the core of RCMSE."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces detrended fluctuation analysis, the monofractal precursor that MF-DFA extends."},{"cited_title":"Entropy 17(5), 3110–3123 (2015)","cited_arxiv_id":null,"evidence_quote":"Documents the undefined-entropy problem of ordinary multiscale entropy at large scales, motivating the choice of RCMSE."},{"cited_title":"Fractal and Frac tional 8(10), 571 (2024)","cited_arxiv_id":null,"evidence_quote":"Provides a prior multifractal comparison across cryptocurrencies and forex markets that this study extends to the four selected assets."}],"review_version":1}