Auto-Research PipelineA durable, human-gated pipeline for computational research

← All examples

Research run963f889ffb68· 24 October 2026

Do volatility-managed factor portfolios survive real-time implementation and a horse race against time-series factor momentum? A century of US market, size and value factors

CompleteArchived runFinance · Modelled on Journal of Finance

Paper

Volatility Management from Monthly Returns versus Factor Momentum: The Fama-French Market, Size, and Value Factors, 1926–2026

Abstract

We test volatility management from monthly returns, not Moreira and Muir's daily signal, on the three Fama-French factors, 1926–2026. Its alphas are indistinguishable from zero with or without a momentum control, and momentum alphas survive a volatility control; of 24 benchmark alphas only the market's twelve-month-window alpha is significant at 5%, though not jointly with size and value (p = 0.10). Real-time annualized Sharpe ratios fall by 0.11 to 0.22 (gains above 0.07 rejected); a variance-forecast weight removes most of this loss, significantly for value (5% level) and size (10%) but not the market, without a detectable gain.

Keywordsvolatility-managed portfoliosfactor momentumtime-series momentumreal-time implementationvariance forecastingSharpe-ratio tests

§Ask about this paper

Grounded in this run

Questions are answered from this run's record — the paper, the executed analysis, the data provenance and the referee report. Nothing else is consulted.

This run is finished, so the recorded paper cannot be rewritten. Questions are still answered from its record.

Pipeline

1.Literature review

Attempt 3Done

Literature review complete (live:agent:claude-opus-5-5): 10 key papers, 41 grounded sources, 5 open gaps.

Summary

Volatility timing has a long pedigree: Fleming, Kirby and Ostdiek (2001, 2003) showed that conditional mean-variance investors gain from timing volatility. Moreira and Muir (2017) turned this into a simple rule - scale a factor by the inverse of its recent realized variance, with a constant that equates the managed and unmanaged volatilities - and found large spanning-regression alphas and higher Sharpe ratios for the market, value, momentum, profitability, investment and other factors, because changes in volatility are not matched by proportional changes in expected returns. Lochstoer and Muir (2022) supply a mechanism (slow-moving volatility beliefs that imply a weak or negative risk-return trade-off), consistent with the mixed sign of the conditional mean-variance relation in French, Schwert and Stambaugh (1987) and Glosten, Jagannathan and Runkle (1993). A second wave asks whether these gains are attainable. Liu, Tang and Zhou (2019) show that the market application uses a look-ahead scaling constant and, once corrected, has very large drawdowns and outperforms only in the financial crisis. Cederburg, O'Doherty, Wang and Yan (2020) find, across 103 strategies, that managed portfolios do not systematically beat their unmanaged counterparts in direct comparisons, and that the spanning-implied combinations are not implementable in real time because the spanning regressions are structurally unstable. Barroso and Detzel (2021) find that after costs only the volatility-managed market survives. The verdict is not settled: DeMiguel, Martin-Utrera and Uppal (2024) and Xu (2026) report real-time, net-of-cost gains for multifactor or refined versions. A separate literature documents time-series momentum (Moskowitz, Ooi and Pedersen 2012) and factor momentum: factors are positively autocorrelated and can be timed on their own prior-year return (Gupta and Kelly 2019; Ehsani and Linnainmaa 2022), although asset-by-asset evidence for time-series momentum is weak (Huang, Li, Wang and Zhou 2020). The two literatures meet in Kim, Tse and Wald (2016), who show that most of time-series momentum's alpha comes from volatility scaling, not from the momentum signal. The converse question is open. Because volatility rises after losses (Glosten, Jagannathan and Runkle 1993; Campbell and Hentschel 1992), an inverse-variance weight is low precisely when a sign-based momentum rule is short, so part of a volatility-managed alpha may be implicit trend timing. We found no paper that runs a two-sided horse race between inverse-variance scaling and sign-only time-series factor momentum on the same factors, both in spanning regressions and in real-time out-of-sample comparisons with interval estimates, over a sample long enough to include the Great Depression, which dominates long-run volatility (Schwert 1989), and the decade after the strategy became public (McLean and Pontiff 2016).

Key papers (10)

  1. Volatility-Managed Portfolios. Alan Moreira and Tyler Muir; Journal of Finance 72(4), 1611-1644, 2017.
    Defines the volatility-managed factor (inverse previous-month realized variance, constant c equating volatilities) and the spanning-regression test; reports large alphas and Sharpe-ratio gains for the market, value and other factors.
  2. On the performance of volatility-managed portfolios. Scott Cederburg, Michael S. O'Doherty, Feifei Wang and Xuemin (Sterling) Yan; Journal of Financial Economics 138(1), 95-117, 2020.
    The real-time critique: across 103 strategies managed portfolios do not systematically beat unmanaged ones in direct comparisons, and spanning-implied combinations fail out of sample because of structural instability.
  3. Volatility-Managed Portfolio: Does It Really Work?. Fang Liu, Xiaoxiao Tang and Guofu Zhou; Journal of Portfolio Management 46(1), 38-51, 2019.
    Shows the market application uses a look-ahead scaling constant; corrected, it has 68%-93% drawdowns and outperforms only during the financial crisis.
  4. Do limits to arbitrage explain the benefits of volatility-managed portfolios?. Pedro Barroso and Andrew Detzel; Journal of Financial Economics 140(3), 744-767, 2021.
    After transaction costs, volatility management of factors other than the market produces zero abnormal returns and lower Sharpe ratios; the managed market is the exception.
  5. A Multifactor Perspective on Volatility-Managed Portfolios. Victor DeMiguel, Alberto Martin-Utrera and Raman Uppal; Journal of Finance 79(6), 3859-3891, 2024.
    A conditional multifactor portfolio outperforms out of sample and net of costs; factor risk prices fall with market volatility - the counterpoint to the single-factor real-time failures.
  6. Improving volatility-managed portfolios in real time. Xia Xu; Critical Finance Review 15(2), 179-207, 2026.
    Most recent real-time evidence: a refined scaling rule yields real-time Sharpe-ratio gains for most of 197 factors and anomalies, robust to leverage limits and costs.
  7. Factor Momentum and the Momentum Factor. Sina Ehsani and Juhani T. Linnainmaa; Journal of Finance 77(3), 1877-1919, 2022.
    Defines time-series factor momentum (long after a positive prior-year factor return, short after a negative one); the average factor earns 6 bp per month after a losing year and 51 bp after a winning year.
  8. Factor Momentum Everywhere. Tarun Gupta and Bryan Kelly; Journal of Portfolio Management 45(3), 13-36, 2019.
    65 factors can be timed on their own recent returns; combined time-series factor momentum earns an annual Sharpe ratio of 0.84.
  9. Time series momentum and volatility scaling. Abby Y. Kim, Yiuman Tse and John K. Wald; Journal of Financial Markets 30, 103-124, 2016.
    Most of time-series momentum's alpha is due to volatility scaling rather than the momentum signal - the direct precedent for separating the two components in a horse race.
  10. Time series momentum. Tobias J. Moskowitz, Yao Hua Ooi and Lasse Heje Pedersen; Journal of Financial Economics 104(2), 228-250, 2012.
    Canonical time-series momentum: the past 12-month return predicts the next month's in 58 futures; positions are volatility-scaled.

Methods in use

  • Spanning regressions of the managed on the unmanaged factor (Huberman and Kandel 1987; Moreira and Muir 2017), reporting annualised alphas, betas, R-squared and appraisal ratios; joint alpha tests in the spirit of Gibbons, Ross and Shanken (1989).
  • Heteroskedasticity- and autocorrelation-consistent (Newey and West 1987) standard errors for monthly return regressions.
  • Direct out-of-sample comparisons of annualised Sharpe ratios with robust tests of the difference: Jobson and Korkie (1981), Lo (2002), and the HAC delta-method or block-bootstrap tests of Ledoit and Wolf (2008).
  • Real-time (expanding-window) estimation of the scaling constant and of spanning-implied combination weights, evaluated out of sample by Sharpe ratio and certainty-equivalent return (Cederburg et al. 2020; Liu, Tang and Zhou 2019), with the estimation-error caution of DeMiguel, Garlappi and Uppal (2009) and the out-of-sample discipline of Goyal and Welch (2008) and Campbell and Thompson (2008).
  • Time-series (factor) momentum rules: sign of the prior 12-month return (Moskowitz, Ooi and Pedersen 2012; Ehsani and Linnainmaa 2022; Gupta and Kelly 2019).
  • Implementation diagnostics: leverage, turnover, maximum drawdown and break-even transaction costs (Barroso and Detzel 2021; Novy-Marx and Velikov 2016); subsample and post-publication splits (McLean and Pontiff 2016).

Open gaps

  • A two-sided horse race between inverse-variance scaling and sign-only time-series factor momentum on the same factors: does the volatility-managed alpha survive controlling for factor momentum, and vice versa?
  • The same horse race in real time: does adding the volatility signal to an expanding-window timing rule that already uses factor momentum raise the out-of-sample Sharpe ratio, and does adding momentum to a volatility rule?
  • Interval estimates (HAC and block-bootstrap) for the Sharpe-ratio differences between managed and unmanaged factors over a century-long sample, rather than point estimates alone.
  • Whether the gains depend on the Great Depression and other high-volatility episodes, and whether they persisted in the decade after the strategy became public (2016 onwards).
  • Implementability with monthly data: leverage caps, turnover and break-even costs of volatility rules measurable from monthly factor returns alone.
Grounded search sources (41) · mode: live:agent:claude-opus-5-5

2.Ideation

Attempt 2Done

3 candidate ideas generated; selected: “Do volatility-managed factor portfolios survive real-time implementation and a horse race against time-series factor momentum? A century of US market, size and value factors”.

  1. Idea 1Selected

    Do volatility-managed factor portfolios survive real-time implementation and a horse race against time-series factor momentum? A century of US market, size and value factors

    Hypothesis. H1 (benchmark): over the full sample, the volatility-managed (VM) version of each factor - the factor scaled by c divided by its trailing realized variance - earns a positive alpha against the unmanaged factor in a Moreira-Muir spanning regression. H2 (real time): when the scaling constant is estimated only from data available at each month, the VM factor's out-of-sample annualised Sharpe ratio over 1947:07-2026:07 exceeds the unmanaged factor's, with a 95% Ledoit-Wolf HAC interval for the difference that excludes zero. H3 (horse race): the VM alpha remains positive at the 5% level when the time-series factor-momentum (TSFM) strategy - long the factor after a positive prior 12-month return, short after a negative one - is added to the spanning regression (H3a), and adding the volatility signal to a real-time timing rule that already uses momentum raises the out-of-sample Sharpe ratio (H3b). Pre-committed decision rule: VM 'survives' for a factor if H2 and H3a both hold; the overall hypothesis is supported only if VM survives for at least two of the three factors. Because volatility rises after losses, the inverse-variance weight is mechanically low when TSFM is short, so the economic alternative is that part of the VM alpha is implicit trend timing.

    Dataset. Fama-French three factors, monthly ff_factors_monthly

    Test plan, novelty and risk

    Test plan. Data: ff_factors_monthly, Mkt-RF, SMB and HML in percent per month, July 1926 to July 2026 (1,201 months); check for missing or sentinel values and report the analysed sample size. Signals known at the end of month t: realized variance RV_t = mean of the squared factor returns of months t-5..t (baseline six-month window); prior-year return m_t = compounded factor return over months t-11..t; TSFM position s_t = +1 if m_t > 0 and -1 otherwise. Strategies earned in month t+1: unmanaged f; VM = (c/RV_t) f, with c equating full-sample volatilities in the in-sample tests and c_t estimated on an expanding window (data through t only) in the real-time tests; TSFM = s_t f. In-sample tests, 1927:07-2026:07 (first month with both signals; 1,189 months): OLS spanning regressions VM = a + b f + e, VM = a + b f + d TSFM + e, TSFM = a + b f + e and TSFM = a + b f + d VM + e, with Newey-West (12-lag) t-statistics, alphas in percent per year and 95% confidence intervals. Real-time tests, out-of-sample period 1947:07-2026:07 (949 months; the first 240 months serve only as the initial estimation window, and every constant, mean, covariance and threshold is re-estimated each month on an expanding window): annualised Sharpe ratios with 95% intervals for f, real-time VM, VM with leverage capped at 2, TSFM, and expanding-window mean-variance combinations {f, VM}, {f, TSFM} and {f, VM, TSFM} (each rescaled to the factor's estimated volatility); Sharpe-ratio differences tested with the Ledoit-Wolf (2008) HAC delta method (Bartlett kernel, 12 lags) and, for the headline difference, a circular block bootstrap (block length 12, 2,000 draws, fixed seed). Mechanism: correlation of the VM weight with the TSFM position; mean return and Sharpe ratio in high- versus low-variance states (RV_t above versus below its expanding-window median) and after positive versus negative prior-year returns, with Newey-West t-statistics for the differences. Implementation: mean and maximum leverage, share of months with leverage above 2, mean monthly turnover |w_t - w_{t-1}|, and the break-even one-way cost per unit of turnover at which the net VM Sharpe ratio equals the unmanaged one. Robustness: 3- and 12-month variance windows, 1/volatility (volatility targeting) instead of 1/variance, the leverage cap, and subperiods 1947:07-1986:12, 1987:01-2026:07 and 2016:01-2026:07 (after the strategy became public). Refutation: H1 fails if the in-sample alpha's 95% interval includes zero; H2 fails if the real-time Sharpe difference is negative or its interval includes zero (called 'refuted' only if the interval lies below zero, otherwise 'inconclusive'); H3a fails if the alpha controlling for TSFM is not significantly positive. Only analyses listed here will be reported.

    Novelty. Kim, Tse and Wald (2016) showed that time-series momentum's alpha is mostly volatility scaling; we ask the converse - how much of the volatility-managed alpha is momentum timing - with a two-sided horse race in which the momentum rule is sign-only (no volatility scaling) and the volatility rule carries no momentum signal. We run the horse race both in spanning regressions and in real time, where Cederburg et al. (2020) and Liu, Tang and Zhou (2019) find the in-sample gains disappear but Xu (2026) and DeMiguel, Martin-Utrera and Uppal (2024) find gains, and we report HAC and bootstrap intervals for every Sharpe difference over a century-long sample that includes the Great Depression and the decade after the strategy became public.

    Expected outcome. A positive in-sample VM alpha for the market factor, weaker for HML and SMB; real-time Sharpe-ratio differences close to zero with wide intervals; positive correlation between the VM weight and the TSFM position, so that controlling for TSFM shrinks the market VM alpha. A plausible overall verdict is that the hypothesis is not supported, with the size of the shortfall and its precision reported factor by factor.

    Main risk. Only monthly factor returns are available, so the variance signal is a six-month trailing realized variance - much noisier and slower than the daily realized variance Moreira and Muir use; negative results speak to monthly-measurable volatility management, not to the daily-data strategy. Inverse-variance weights can imply extreme leverage in calm periods, returns are gross of costs, Sharpe-ratio differences have low power, the pre-1963 factor data rest on thinner coverage, and a few high-volatility episodes (1929-1939, 2008) can dominate the results.

  2. Idea 2

    Volatility management across the 25 size and book-to-market portfolios: a cross-section of test assets for the real-time and spanning tests

    Hypothesis. Inverse-variance scaling of the 25 size and book-to-market portfolios produces a jointly non-zero set of alphas relative to the unmanaged portfolios (GRS test rejects at 5%), but in real time the average out-of-sample Sharpe-ratio gain across the 25 portfolios is not positive.

    Dataset. Fama-French 25 portfolios formed on size and book-to-market, monthly value-weighted returns ff_25_portfolios_5x5

    Test plan, novelty and risk

    Test plan. Data: ff_25_portfolios_5x5, 25 value-weighted portfolio total returns in percent per month, 1926:07-2026:07; excess returns are not available in this file without the risk-free rate, so the test is run on total returns and on long-short spreads (small minus big within each book-to-market quintile). Signals: six-month trailing realized variance per portfolio. Estimate 25 spanning regressions with Newey-West (12-lag) errors and a GRS joint test; compute real-time Sharpe-ratio differences with an expanding-window scaling constant over 1947:07-2026:07, with Ledoit-Wolf HAC intervals and a cross-sectional average. Refuted if the GRS test does not reject or if the average real-time gain is significantly positive.

    Novelty. Raises power relative to three factors by using 25 test assets, which the factor-level studies cannot.

    Expected outcome. In-sample alphas concentrated in small-cap portfolios; real-time gains near zero on average.

    Main risk. Total rather than excess returns (no risk-free rate in this dataset), heavy cross-sectional correlation that limits the effective number of tests, and small-cap illiquidity before 1963.

  3. Idea 3

    Is the market's conditional risk-return trade-off flat? Real-time variance states and the next-month equity premium, 1926-2026

    Hypothesis. The next-month market excess return does not rise in proportion to the trailing realized variance: the conditional Sharpe ratio in high-variance months (above the expanding-window median) is lower than in low-variance months, and a predictive regression of Mkt-RF on lagged realized variance has a slope below the proportional benchmark implied by a constant price of risk.

    Dataset. Fama-French three factors, monthly ff_factors_monthly

    Test plan, novelty and risk

    Test plan. Data: ff_factors_monthly, Mkt-RF in percent per month, 1927:01-2026:07. Real-time variance states from an expanding-window median of six-month realized variance; predictive regression of Mkt-RF(t+1) on RV(t) with Newey-West (12-lag) errors; out-of-sample R-squared relative to the historical mean (Campbell-Thompson) with an expanding window from 1947. Refuted if the high-variance-state Sharpe ratio exceeds the low-variance one with a 95% interval excluding zero.

    Novelty. Revisits French, Schwert and Stambaugh (1987) and Glosten, Jagannathan and Runkle (1993) with a century of data and real-time state definitions.

    Expected outcome. A positive but less-than-proportional relation, with wide intervals.

    Main risk. Monthly realized variance is a noisy proxy for conditional variance; the result is a restatement of an old debate rather than a new mechanism.

Ranking rationale

Idea 0 answers the research question directly with the dataset that carries a century of the three factors, and its two-sided horse race plus real-time evaluation addresses the open gaps; idea 1 adds test assets but loses excess returns and the momentum comparison; idea 2 isolates one mechanism behind idea 0 and is less novel.

3.Experimentation

Attempt 6Done

Experiment succeeded in 1 round (1.15s).

Reviewer feedback applied: Independent referee (Claude Fable 5.1), round 2: minor_revision. Required changes: (1) [analysis] Add one row to Table VII Panel A: a managed factor with weight proportional to 1/(a_t + b_t RV_t), where a_t and b_t are the expanding-window estimates of equation (8) through month t, with its real-time SR difference and interval; discuss it in IV.D as the direct test of the noisy-signal explanation. If it is not run, then [text] change 'the evidence points to the signal' (VI) and 'the signal is an over-dispersed, noisy forecast' (IV.D) to wording that says the forecast slopes are consistent with a noisy signal and that a forecast-based weight was not tested. (2) [text] Move Haddad, Kozak, and Santosh (2020) out of the list of papers that time factors by their own past returns (Section I, third paragraph) and cite it for factor timing in general, for example in the sentence that introduces the factor-timing literature. (3) [text] Add one sentence, in IV.C after 'H2 fails for every factor' or in VI, stating the overall verdict under the pre-committed rule: VM survives for none of the three factors, so the hypothesis is not supported; the narrower proposition that real-time Sharpe ratios are equal is left inconclusive because no interval lies entirely below zero. (4) [text] In the notes to Table IV, state that the stars in Panel C test the slope against zero and that the 't for slope = 1' column tests it against one; in IV.D replace 'that variant also loses (Table VII)' with the point estimates (-0.05, -0.10, -0.07) and the statement that none is significant at the 5% level. (5) [analysis] Editor's request, for consistency with the companion Chinese edition (same design, script and data): report the plain H1 alpha (VM on f, Newey-West 12-lag 95% interval) under every alternative specification of the robustness table (3- and 12-month RV, 1/sqrt(RV), cap 2, the subperiods), with the HAC Wald joint test across the three factors for each, and qualify H1 accordingly; the Chinese edition finds the 12-month-window market alpha significant at 5% (3.35% a year [0.26, 6.44]; joint p = 0.10).
Succeeded1 round · exit 0 · 1.15s

Data provenance

Datasetff_factors_monthly

Fama-French three factors, monthly

Source
Kenneth R. French Data Library, Tuck School of Business, Dartmouth College
Citation
Fama, E.F. and French, K.R. (1993) Common risk factors in the returns on stocks and bonds. Journal of Financial Economics 33(1), 3-56. Data: Kenneth R. French Data Library, Tuck School of Business, Dartmouth College.
Licence
Copyright Eugene F. Fama and Kenneth R. French. No formal licence is posted; the library is distributed for research use and is universally cited in the academic literature. Attribution to the Data Library is required.
File
https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/ftp/F-F_Research_Data_Factors_CSV.zip
Observations
1,201
Variables (5)
Date, Mkt-RF, SMB, HML, RF
Units
Mkt-RF, SMB and HML are excess/long-short returns in percent per month (divide by 100 for decimals). RF is the one-month Treasury bill rate, also percent per month. Date is the first day of the month.
Retrieved
24 Oct 2026, 02:10
Checksum
sha256 b840dba55d319f48…

Real, publicly available data — downloaded and checksum-verified before the analysis ran, never simulated.

Result (RESULT_JSON)

{
  "sample": {
    "n_raw": 1201,
    "n_missing": 0,
    "calendar_gaps": 0,
    "first": "1926-07",
    "last": "2026-07",
    "insample": "1927-07 to 2026-07",
    "oos": "1947-07 to 2026-07",
    "n_insample": 1189,
    "n_oos": 949,
    "n_2016_2026": 127
  },
  "settings": {
    "nw_lags": 12,
    "rv_window": 6,
    "leverage_cap": 2,
    "boot_draws": 2000,
    "boot_block": 12,
    "seed": 20261024
  },
  "summary": {
    "Mkt-RF": {
      "mean_pa": 8.2556,
      "sd_pa": 18.4435,
      "sr": {
        "est": 0.4478,
        "lo": 0.2185,
        "hi": 0.6771,
        "p": 0.0001
      },
      "sr_vm": {
        "est": 0.3818,
        "lo": 0.1477,
        "hi": 0.6159,
        "p": 0.0014
      },
      "sr_ts": {
        "est": 0.3177,
        "lo": 0.1214,
        "hi": 0.5139,
        "p": 0.0015
      },
      "share_long": 0.7065,
      "c_full": 12.7505,
      "corr_w_s": 0.2614,
      "w_is_long": 1.273,
      "w_is_short": 0.682,
      "w_is_max": 11.7137,
      "share_w_is_gt2": 0.1262
    },
    "SMB": {
      "mean_pa": 2.0668,
      "sd_pa": 10.9457,
      "sr": {
        "est": 0.1889,
        "lo": -0.0142,
        "hi": 0.392,
        "p": 0.0683
      },
      "sr_vm": {
        "est": 0.0028,
        "lo": -0.2153,
        "hi": 0.2209,
        "p": 0.9802
      },
      "sr_ts": {
        "est": 0.2981,
        "lo": 0.0981,
        "hi": 0.4981,
        "p": 0.0035
      },
      "share_long": 0.529,
      "c_full": 3.5608,
      "corr_w_s": -0.1676,
      "w_is_long": 0.813,
      "w_is_short": 1.254,
      "w_is_max": 17.9444,
      "share_w_is_gt2": 0.1194
    },
    "HML": {
      "mean_pa": 4.1872,
      "sd_pa": 12.332,
      "sr": {
        "est": 0.3397,
        "lo": 0.1319,
        "hi": 0.5475,
        "p": 0.0014
      },
      "sr_vm": {
        "est": 0.1657,
        "lo": -0.0388,
        "hi": 0.3701,
        "p": 0.1122
      },
      "sr_ts": {
        "est": 0.3017,
        "lo": 0.1085,
        "hi": 0.495,
        "p": 0.0022
      },
      "share_long": 0.6114,
      "c_full": 3.3365,
      "corr_w_s": 0.0121,
      "w_is_long": 1.0511,
      "w_is_short": 1.0145,
      "w_is_max": 29.0044,
      "share_w_is_gt2": 0.1211
    }
  },
  "mechanism": {
    "Mkt-RF": {
      "high_vs_low_var": {
        "mean1": 0.8786,
        "mean0": 0.5585,
        "sd1": 6.699,
        "sd0": 4.1379,
        "sr1": 0.4548,
        "sr0": 0.4679,
        "diff": 0.3201,
        "diff_t": 0.9078,
        "diff_lo": -0.371,
        "diff_hi": 1.0112,
        "n1": 481,
        "n0": 708
      },
      "pos_vs_neg_mom": {
        "mean1": 0.8337,
        "mean0": 0.3372,
        "sd1": 4.3163,
        "sd0": 7.1888,
        "sr1": 0.6695,
        "sr0": 0.1627,
        "diff": 0.4965,
        "diff_t": 1.2607,
        "diff_lo": -0.2754,
        "diff_hi": 1.2684,
        "n1": 840,
        "n0": 349
      }
    },
    "SMB": {
      "high_vs_low_var": {
        "mean1": 0.3014,
        "mean0": 0.0451,
        "sd1": 3.8752,
        "sd0": 2.2384,
        "sr1": 0.2696,
        "sr0": 0.0698,
        "diff": 0.2563,
        "diff_t": 1.3952,
        "diff_lo": -0.1038,
        "diff_hi": 0.6164,
        "n1": 590,
        "n0": 599
      },
      "pos_vs_neg_mom": {
        "mean1": 0.4191,
        "mean0": -0.1051,
        "sd1": 3.412,
        "sd0": 2.8274,
        "sr1": 0.4259,
        "sr0": -0.1288,
        "diff": 0.5242,
        "diff_t": 2.8177,
        "diff_lo": 0.1596,
        "diff_hi": 0.8888,
        "n1": 629,
        "n0": 560
      }
    },
    "HML": {
      "high_vs_low_var": {
        "mean1": 0.5126,
        "mean0": 0.2213,
        "sd1": 4.6897,
        "sd0": 2.3214,
        "sr1": 0.379,
        "sr0": 0.3305,
        "diff": 0.2913,
        "diff_t": 1.2397,
        "diff_lo": -0.1693,
        "diff_hi": 0.7519,
        "n1": 521,
        "n0": 668
      },
      "pos_vs_neg_mom": {
        "mean1": 0.5391,
        "mean0": 0.0497,
        "sd1": 3.3882,
        "sd0": 3.7991,
        "sr1": 0.5515,
        "sr0": 0.0454,
        "diff": 0.4893,
        "diff_t": 2.4647,
        "diff_lo": 0.1002,
        "diff_hi": 0.8784,
        "n1": 727,
        "n0": 462
      }
    }
  },
  "spanning": {
    "Mkt-RF": {
      "vm_f": {
        "alpha": 1.6438,
        "t": 0.9049,
        "lo": -1.9166,
        "hi": 5.2043,
        "p": 0.3655,
        "beta": [
          0.6535
        ],
        "beta_t": [
          6.3931
        ],
        "r2_adj": 0.4266,
        "n": 1189
      },
      "vm_f_ts": {
        "alpha": -0.3647,
        "t": -0.2318,
        "lo": -3.4484,
        "hi": 2.7189,
        "p": 0.8167,
        "beta": [
          0.674,
          0.3128
        ],
        "beta_t": [
          9.4868,
          5.6356
        ],
        "r2_adj": 0.5245,
        "n": 1189
      },
      "ts_f": {
        "alpha": 6.4219,
        "t": 2.966,
        "lo": 2.1782,
        "hi": 10.6655,
        "p": 0.003,
        "beta": [
          -0.0656
        ],
        "beta_t": [
          -0.4395
        ],
        "r2_adj": 0.0034,
        "n": 1189
      },
      "ts_f_vm": {
        "alpha": 5.5209,
        "t": 3.1313,
        "lo": 2.0652,
        "hi": 8.9766,
        "p": 0.0017,
        "beta": [
          -0.4238,
          0.5481
        ],
        "beta_t": [
          -2.5917,
          5.3375
        ],
        "r2_adj": 0.1736,
        "n": 1189
      }
    },
    "SMB": {
      "vm_f": {
        "alpha": -1.1666,
        "t": -1.2455,
        "lo": -3.0025,
        "hi": 0.6692,
        "p": 0.2129,
        "beta": [
          0.579
        ],
        "beta_t": [
          7.3745
        ],
        "r2_adj": 0.3347,
        "n": 1189
      },
      "vm_f_ts": {
        "alpha": -0.8084,
        "t": -0.8584,
        "lo": -2.6541,
        "hi": 1.0373,
        "p": 0.3907,
        "beta": [
          0.6108,
          -0.1302
        ],
        "beta_t": [
          9.0605,
          -2.0707
        ],
        "r2_adj": 0.3501,
        "n": 1189
      },
      "ts_f": {
        "alpha": 2.7505,
        "t": 2.6126,
        "lo": 0.687,
        "hi": 4.8141,
        "p": 0.009,
        "beta": [
          0.2437
        ],
        "beta_t": [
          1.6245
        ],
        "r2_adj": 0.0589,
        "n": 1189
      },
      "ts_f_vm": {
        "alpha": 2.5365,
        "t": 2.3822,
        "lo": 0.4496,
        "hi": 4.6235,
        "p": 0.0172,
        "beta": [
          0.35,
          -0.1834
        ],
        "beta_t": [
          1.912,
          -2.0029
        ],
        "r2_adj": 0.0806,
        "n": 1189
      }
    },
    "HML": {
      "vm_f": {
        "alpha": 0.1401,
        "t": 0.1308,
        "lo": -1.9589,
        "hi": 2.2391,
        "p": 0.8959,
        "beta": [
          0.4542
        ],
        "beta_t": [
          4.5821
        ],
        "r2_adj": 0.2056,
        "n": 1189
      },
      "vm_f_ts": {
        "alpha": 0.1944,
        "t": 0.1814,
        "lo": -1.906,
        "hi": 2.2947,
        "p": 0.8561,
        "beta": [
          0.4562,
          -0.0168
        ],
        "beta_t": [
          4.6642,
          -0.3034
        ],
        "r2_adj": 0.2052,
        "n": 1189
      },
      "ts_f": {
        "alpha": 3.2334,
        "t": 2.864,
        "lo": 1.0205,
        "hi": 5.4462,
        "p": 0.0042,
        "beta": [
          0.117
        ],
        "beta_t": [
          0.9356
        ],
        "r2_adj": 0.0128,
        "n": 1189
      },
      "ts_f_vm": {
        "alpha": 3.2363,
        "t": 2.8704,
        "lo": 1.0264,
        "hi": 5.4462,
        "p": 0.0041,
        "beta": [
          0.1265,
          -0.0209
        ],
        "beta_t": [
          0.8632,
          -0.2969
        ],
        "r2_adj": 0.0123,
        "n": 1189
      }
    }
  },
  "realtime": {
    "Mkt-RF": {
      "sr": {
        "f": {
          "est": 0.5573,
          "lo": 0.3035,
          "hi": 0.8112,
          "p": 0
        },
        "vm_rt": {
          "est": 0.4437,
          "lo": 0.1944,
          "hi": 0.6929,
          "p": 0.0005
        },
        "vm_cap": {
          "est": 0.4697,
          "lo": 0.2054,
          "hi": 0.734,
          "p": 0.0005
        },
        "vm_expost": {
          "est": 0.4231,
          "lo": 0.1738,
          "hi": 0.6724,
          "p": 0.0009
        },
        "ts": {
          "est": 0.3379,
          "lo": 0.1229,
          "hi": 0.5529,
          "p": 0.0021
        },
        "c_fvm": {
          "est": 0.5129,
          "lo": 0.2617,
          "hi": 0.7641,
          "p": 0.0001
        },
        "c_fts": {
          "est": 0.5908,
          "lo": 0.3666,
          "hi": 0.8151,
          "p": 0
        },
        "c_all": {
          "est": 0.558,
          "lo": 0.3357,
          "hi": 0.7803,
          "p": 0
        }
      },
      "dsr": {
        "vm_rt-f": {
          "est": -0.1137,
          "lo": -0.2939,
          "hi": 0.0666,
          "p": 0.2164
        },
        "vm_cap-f": {
          "est": -0.0877,
          "lo": -0.2307,
          "hi": 0.0554,
          "p": 0.2296
        },
        "vm_expost-f": {
          "est": -0.1343,
          "lo": -0.314,
          "hi": 0.0455,
          "p": 0.1433
        },
        "ts-f": {
          "est": -0.2194,
          "lo": -0.5155,
          "hi": 0.0766,
          "p": 0.1463
        },
        "c_fvm-f": {
          "est": -0.0445,
          "lo": -0.1497,
          "hi": 0.0608,
          "p": 0.4079
        },
        "c_fts-f": {
          "est": 0.0335,
          "lo": -0.1231,
          "hi": 0.1901,
          "p": 0.6751
        },
        "c_all-f": {
          "est": 0.0006,
          "lo": -0.1581,
          "hi": 0.1594,
          "p": 0.9937
        },
        "c_all-c_fts": {
          "est": -0.0329,
          "lo": -0.0794,
          "hi": 0.0137,
          "p": 0.1667
        },
        "c_all-c_fvm": {
          "est": 0.0451,
          "lo": -0.0887,
          "hi": 0.1789,
          "p": 0.5089
        }
      },
      "boot_vm_rt-f": [
        -0.2871,
        0.0651
      ],
      "n": 949
    },
    "SMB": {
      "sr": {
        "f": {
          "est": 0.1167,
          "lo": -0.1194,
          "hi": 0.3527,
          "p": 0.3326
        },
        "vm_rt": {
          "est": -0.0703,
          "lo": -0.3166,
          "hi": 0.1759,
          "p": 0.5756
        },
        "vm_cap": {
          "est": -0.0577,
          "lo": -0.2978,
          "hi": 0.1824,
          "p": 0.6375
        },
        "vm_expost": {
          "est": -0.0516,
          "lo": -0.2978,
          "hi": 0.1946,
          "p": 0.6811
        },
        "ts": {
          "est": 0.2601,
          "lo": 0.0366,
          "hi": 0.4837,
          "p": 0.0226
        },
        "c_fvm": {
          "est": 0.1066,
          "lo": -0.1318,
          "hi": 0.3449,
          "p": 0.381
        },
        "c_fts": {
          "est": 0.2441,
          "lo": 0.0155,
          "hi": 0.4726,
          "p": 0.0363
        },
        "c_all": {
          "est": 0.2103,
          "lo": -0.0183,
          "hi": 0.4388,
          "p": 0.0713
        }
      },
      "dsr": {
        "vm_rt-f": {
          "est": -0.187,
          "lo": -0.41,
          "hi": 0.036,
          "p": 0.1003
        },
        "vm_cap-f": {
          "est": -0.1744,
          "lo": -0.3356,
          "hi": -0.0132,
          "p": 0.0339
        },
        "vm_expost-f": {
          "est": -0.1683,
          "lo": -0.38,
          "hi": 0.0434,
          "p": 0.1193
        },
        "ts-f": {
          "est": 0.1434,
          "lo": -0.1458,
          "hi": 0.4327,
          "p": 0.3311
        },
        "c_fvm-f": {
          "est": -0.0101,
          "lo": -0.1995,
          "hi": 0.1793,
          "p": 0.9165
        },
        "c_fts-f": {
          "est": 0.1274,
          "lo": -0.1034,
          "hi": 0.3581,
          "p": 0.2792
        },
        "c_all-f": {
          "est": 0.0936,
          "lo": -0.1634,
          "hi": 0.3506,
          "p": 0.4754
        },
        "c_all-c_fts": {
          "est": -0.0338,
          "lo": -0.1548,
          "hi": 0.0873,
          "p": 0.5844
        },
        "c_all-c_fvm": {
          "est": 0.1037,
          "lo": -0.0873,
          "hi": 0.2947,
          "p": 0.2872
        }
      },
      "boot_vm_rt-f": [
        -0.4107,
        0.0222
      ],
      "n": 949
    },
    "HML": {
      "sr": {
        "f": {
          "est": 0.3779,
          "lo": 0.1139,
          "hi": 0.642,
          "p": 0.005
        },
        "vm_rt": {
          "est": 0.1579,
          "lo": -0.0648,
          "hi": 0.3807,
          "p": 0.1646
        },
        "vm_cap": {
          "est": 0.274,
          "lo": 0.0204,
          "hi": 0.5277,
          "p": 0.0342
        },
        "vm_expost": {
          "est": 0.1478,
          "lo": -0.0886,
          "hi": 0.3842,
          "p": 0.2205
        },
        "ts": {
          "est": 0.3053,
          "lo": 0.0704,
          "hi": 0.5403,
          "p": 0.0109
        },
        "c_fvm": {
          "est": 0.3066,
          "lo": 0.058,
          "hi": 0.5552,
          "p": 0.0157
        },
        "c_fts": {
          "est": 0.4446,
          "lo": 0.2145,
          "hi": 0.6747,
          "p": 0.0002
        },
        "c_all": {
          "est": 0.3821,
          "lo": 0.1443,
          "hi": 0.6198,
          "p": 0.0016
        }
      },
      "dsr": {
        "vm_rt-f": {
          "est": -0.22,
          "lo": -0.4515,
          "hi": 0.0115,
          "p": 0.0625
        },
        "vm_cap-f": {
          "est": -0.1039,
          "lo": -0.2974,
          "hi": 0.0897,
          "p": 0.2928
        },
        "vm_expost-f": {
          "est": -0.2301,
          "lo": -0.4588,
          "hi": -0.0015,
          "p": 0.0485
        },
        "ts-f": {
          "est": -0.0726,
          "lo": -0.4089,
          "hi": 0.2638,
          "p": 0.6724
        },
        "c_fvm-f": {
          "est": -0.0713,
          "lo": -0.1723,
          "hi": 0.0296,
          "p": 0.1662
        },
        "c_fts-f": {
          "est": 0.0667,
          "lo": -0.0993,
          "hi": 0.2327,
          "p": 0.4309
        },
        "c_all-f": {
          "est": 0.0042,
          "lo": -0.1708,
          "hi": 0.1791,
          "p": 0.9629
        },
        "c_all-c_fts": {
          "est": -0.0625,
          "lo": -0.1539,
          "hi": 0.0288,
          "p": 0.1796
        },
        "c_all-c_fvm": {
          "est": 0.0755,
          "lo": -0.0508,
          "hi": 0.2017,
          "p": 0.2414
        }
      },
      "boot_vm_rt-f": [
        -0.4425,
        0.015
      ],
      "n": 949
    }
  },
  "implementation": {
    "Mkt-RF": {
      "vm_rt": {
        "mean_w": 1.4716,
        "max_w": 12.0962,
        "share_w_gt2": 0.216,
        "turnover": 0.37,
        "breakeven_bp": null
      },
      "vm_cap": {
        "mean_w": 1.177,
        "max_w": 2,
        "share_w_gt2": 0,
        "turnover": 0.1639,
        "breakeven_bp": null
      },
      "ts": {
        "mean_w": 0.4647,
        "max_w": 1,
        "share_w_gt2": 0,
        "turnover": 0.1456,
        "breakeven_bp": null
      },
      "c_fvm": {
        "mean_w": 1.3406,
        "max_w": 7.1353,
        "share_w_gt2": 0.0959,
        "turnover": 0.1658,
        "breakeven_bp": null
      },
      "c_fts": {
        "mean_w": 1.1678,
        "max_w": 1.7162,
        "share_w_gt2": 0,
        "turnover": 0.1083,
        "breakeven_bp": 46.4454
      },
      "c_all": {
        "mean_w": 1.1751,
        "max_w": 4.0872,
        "share_w_gt2": 0.0306,
        "turnover": 0.1669,
        "breakeven_bp": 0.5895
      }
    },
    "SMB": {
      "vm_rt": {
        "mean_w": 1.2569,
        "max_w": 18.4601,
        "share_w_gt2": 0.1823,
        "turnover": 0.3594,
        "breakeven_bp": null
      },
      "vm_cap": {
        "mean_w": 0.9268,
        "max_w": 2,
        "share_w_gt2": 0,
        "turnover": 0.1542,
        "breakeven_bp": null
      },
      "ts": {
        "mean_w": -0.0263,
        "max_w": 1,
        "share_w_gt2": 0,
        "turnover": 0.2131,
        "breakeven_bp": 53.9205
      },
      "c_fvm": {
        "mean_w": 0.4999,
        "max_w": 16.0149,
        "share_w_gt2": 0.0211,
        "turnover": 0.2099,
        "breakeven_bp": null
      },
      "c_fts": {
        "mean_w": 0.3367,
        "max_w": 1.2302,
        "share_w_gt2": 0,
        "turnover": 0.1882,
        "breakeven_bp": 52.564
      },
      "c_all": {
        "mean_w": 0.1434,
        "max_w": 11.5131,
        "share_w_gt2": 0.0211,
        "turnover": 0.2731,
        "breakeven_bp": 28.7046
      }
    },
    "HML": {
      "vm_rt": {
        "mean_w": 1.3219,
        "max_w": 37.9536,
        "share_w_gt2": 0.1665,
        "turnover": 0.4637,
        "breakeven_bp": null
      },
      "vm_cap": {
        "mean_w": 0.9477,
        "max_w": 2,
        "share_w_gt2": 0,
        "turnover": 0.1711,
        "breakeven_bp": null
      },
      "ts": {
        "mean_w": 0.2308,
        "max_w": 1,
        "share_w_gt2": 0,
        "turnover": 0.1857,
        "breakeven_bp": null
      },
      "c_fvm": {
        "mean_w": 1.297,
        "max_w": 15.4356,
        "share_w_gt2": 0.0843,
        "turnover": 0.1519,
        "breakeven_bp": null
      },
      "c_fts": {
        "mean_w": 0.8614,
        "max_w": 1.3592,
        "share_w_gt2": 0,
        "turnover": 0.1145,
        "breakeven_bp": 47.3752
      },
      "c_all": {
        "mean_w": 1.1266,
        "max_w": 14.2774,
        "share_w_gt2": 0.0738,
        "turnover": 0.2272,
        "breakeven_bp": 1.726
      }
    }
  },
  "robustness": {
    "Mkt-RF": {
      "baseline": {
        "dsr": {
          "est": -0.1137,
          "lo": -0.2939,
          "hi": 0.0666,
          "p": 0.2164
        },
        "n_rt": 949,
        "alpha_ts": {
          "alpha": -0.3647,
          "lo": -3.4484,
          "hi": 2.7189,
          "t": -0.2318
        },
        "n_is": 1189,
        "alpha_f": {
          "alpha": 1.6438,
          "lo": -1.9166,
          "hi": 5.2043,
          "t": 0.9049,
          "p": 0.3655
        }
      },
      "rv3": {
        "dsr": {
          "est": -0.2746,
          "lo": -0.4752,
          "hi": -0.0739,
          "p": 0.0073
        },
        "n_rt": 949,
        "alpha_ts": {
          "alpha": 0.124,
          "lo": -3.3844,
          "hi": 3.6324,
          "t": 0.0693
        },
        "n_is": 1189,
        "alpha_f": {
          "alpha": 1.3114,
          "lo": -2.3356,
          "hi": 4.9584,
          "t": 0.7048,
          "p": 0.481
        }
      },
      "rv12": {
        "dsr": {
          "est": -0.0543,
          "lo": -0.2093,
          "hi": 0.1006,
          "p": 0.492
        },
        "n_rt": 949,
        "alpha_ts": {
          "alpha": 1.6417,
          "lo": -0.9299,
          "hi": 4.2132,
          "t": 1.2512
        },
        "n_is": 1189,
        "alpha_f": {
          "alpha": 3.3493,
          "lo": 0.2554,
          "hi": 6.4432,
          "t": 2.1218,
          "p": 0.0339
        }
      },
      "inv_vol": {
        "dsr": {
          "est": -0.0452,
          "lo": -0.1453,
          "hi": 0.055,
          "p": 0.3771
        },
        "n_rt": 949,
        "alpha_ts": {
          "alpha": -0.2717,
          "lo": -2.1012,
          "hi": 1.5579,
          "t": -0.291
        },
        "n_is": 1189,
        "alpha_f": {
          "alpha": 1.2459,
          "lo": -1.0564,
          "hi": 3.5481,
          "t": 1.0607,
          "p": 0.2888
        }
      },
      "cap2": {
        "dsr": {
          "est": -0.0877,
          "lo": -0.2307,
          "hi": 0.0554,
          "p": 0.2296
        },
        "n_rt": 949,
        "alpha_ts": {
          "alpha": -0.7385,
          "lo": -3.0154,
          "hi": 1.5385,
          "t": -0.6357
        },
        "n_is": 1189,
        "alpha_f": {
          "alpha": 0.9588,
          "lo": -1.7682,
          "hi": 3.6859,
          "t": 0.6892,
          "p": 0.4907
        }
      },
      "1947_1986": {
        "dsr": {
          "est": -0.1382,
          "lo": -0.3634,
          "hi": 0.0869,
          "p": 0.2288
        },
        "n_rt": 474,
        "alpha_ts": {
          "alpha": -2.2686,
          "lo": -5.908,
          "hi": 1.3708,
          "t": -1.2218
        },
        "n_is": 474,
        "alpha_f": {
          "alpha": -1.4234,
          "lo": -5.306,
          "hi": 2.4593,
          "t": -0.7185,
          "p": 0.4724
        }
      },
      "1987_2026": {
        "dsr": {
          "est": -0.0751,
          "lo": -0.3543,
          "hi": 0.2041,
          "p": 0.5981
        },
        "n_rt": 475,
        "alpha_ts": {
          "alpha": 0.4228,
          "lo": -4.2121,
          "hi": 5.0577,
          "t": 0.1788
        },
        "n_is": 475,
        "alpha_f": {
          "alpha": 1.6357,
          "lo": -3.6905,
          "hi": 6.9619,
          "t": 0.6019,
          "p": 0.5472
        }
      },
      "2016_2026": {
        "dsr": {
          "est": -0.2121,
          "lo": -0.8,
          "hi": 0.3759,
          "p": 0.4796
        },
        "n_rt": 127,
        "alpha_ts": {
          "alpha": 2.0447,
          "lo": -10.6185,
          "hi": 14.7079,
          "t": 0.3165
        },
        "n_is": 127,
        "alpha_f": {
          "alpha": 1.268,
          "lo": -12.6679,
          "hi": 15.2039,
          "t": 0.1783,
          "p": 0.8585
        }
      }
    },
    "SMB": {
      "baseline": {
        "dsr": {
          "est": -0.187,
          "lo": -0.41,
          "hi": 0.036,
          "p": 0.1003
        },
        "n_rt": 949,
        "alpha_ts": {
          "alpha": -0.8084,
          "lo": -2.6541,
          "hi": 1.0373,
          "t": -0.8584
        },
        "n_is": 1189,
        "alpha_f": {
          "alpha": -1.1666,
          "lo": -3.0025,
          "hi": 0.6692,
          "t": -1.2455,
          "p": 0.2129
        }
      },
      "rv3": {
        "dsr": {
          "est": -0.0113,
          "lo": -0.2728,
          "hi": 0.2502,
          "p": 0.9326
        },
        "n_rt": 949,
        "alpha_ts": {
          "alpha": 0.9399,
          "lo": -0.8562,
          "hi": 2.736,
          "t": 1.0256
        },
        "n_is": 1189,
        "alpha_f": {
          "alpha": 0.9635,
          "lo": -0.9835,
          "hi": 2.9106,
          "t": 0.97,
          "p": 0.3321
        }
      },
      "rv12": {
        "dsr": {
          "est": -0.1452,
          "lo": -0.324,
          "hi": 0.0336,
          "p": 0.1115
        },
        "n_rt": 949,
        "alpha_ts": {
          "alpha": -0.2929,
          "lo": -1.852,
          "hi": 1.2662,
          "t": -0.3682
        },
        "n_is": 1189,
        "alpha_f": {
          "alpha": -0.7892,
          "lo": -2.3795,
          "hi": 0.8012,
          "t": -0.9726,
          "p": 0.3308
        }
      },
      "inv_vol": {
        "dsr": {
          "est": -0.1033,
          "lo": -0.2144,
          "hi": 0.0078,
          "p": 0.0684
        },
        "n_rt": 949,
        "alpha_ts": {
          "alpha": -0.3916,
          "lo": -1.447,
          "hi": 0.6638,
          "t": -0.7273
        },
        "n_is": 1189,
        "alpha_f": {
          "alpha": -0.6749,
          "lo": -1.7122,
          "hi": 0.3624,
          "t": -1.2752,
          "p": 0.2023
        }
      },
      "cap2": {
        "dsr": {
          "est": -0.1744,
          "lo": -0.3356,
          "hi": -0.0132,
          "p": 0.0339
        },
        "n_rt": 949,
        "alpha_ts": {
          "alpha": -0.5802,
          "lo": -1.6589,
          "hi": 0.4985,
          "t": -1.0542
        },
        "n_is": 1189,
        "alpha_f": {
          "alpha": -0.8348,
          "lo": -1.9429,
          "hi": 0.2733,
          "t": -1.4766,
          "p": 0.1398
        }
      },
      "1947_1986": {
        "dsr": {
          "est": -0.3189,
          "lo": -0.6834,
          "hi": 0.0457,
          "p": 0.0865
        },
        "n_rt": 474,
        "alpha_ts": {
          "alpha": -1.6305,
          "lo": -5.7273,
          "hi": 2.4663,
          "t": -0.7801
        },
        "n_is": 474,
        "alpha_f": {
          "alpha": -2.3849,
          "lo": -6.2547,
          "hi": 1.485,
          "t": -1.2079,
          "p": 0.2271
        }
      },
      "1987_2026": {
        "dsr": {
          "est": -0.0848,
          "lo": -0.2957,
          "hi": 0.126,
          "p": 0.4303
        },
        "n_rt": 475,
        "alpha_ts": {
          "alpha": -0.8014,
          "lo": -2.6209,
          "hi": 1.018,
          "t": -0.8634
        },
        "n_is": 475,
        "alpha_f": {
          "alpha": -0.7543,
          "lo": -2.6141,
          "hi": 1.1056,
          "t": -0.7949,
          "p": 0.4267
        }
      },
      "2016_2026": {
        "dsr": {
          "est": 0.1304,
          "lo": -0.2082,
          "hi": 0.469,
          "p": 0.4502
        },
        "n_rt": 127,
        "alpha_ts": {
          "alpha": 0.7347,
          "lo": -1.5703,
          "hi": 3.0397,
          "t": 0.6247
        },
        "n_is": 127,
        "alpha_f": {
          "alpha": 0.7482,
          "lo": -1.4705,
          "hi": 2.9668,
          "t": 0.6609,
          "p": 0.5087
        }
      }
    },
    "HML": {
      "baseline": {
        "dsr": {
          "est": -0.22,
          "lo": -0.4515,
          "hi": 0.0115,
          "p": 0.0625
        },
        "n_rt": 949,
        "alpha_ts": {
          "alpha": 0.1944,
          "lo": -1.906,
          "hi": 2.2947,
          "t": 0.1814
        },
        "n_is": 1189,
        "alpha_f": {
          "alpha": 0.1401,
          "lo": -1.9589,
          "hi": 2.2391,
          "t": 0.1308,
          "p": 0.8959
        }
      },
      "rv3": {
        "dsr": {
          "est": -0.3629,
          "lo": -0.6578,
          "hi": -0.0679,
          "p": 0.0159
        },
        "n_rt": 949,
        "alpha_ts": {
          "alpha": -0.7678,
          "lo": -3.1863,
          "hi": 1.6508,
          "t": -0.6222
        },
        "n_is": 1189,
        "alpha_f": {
          "alpha": -0.7521,
          "lo": -3.2309,
          "hi": 1.7267,
          "t": -0.5947,
          "p": 0.552
        }
      },
      "rv12": {
        "dsr": {
          "est": -0.1375,
          "lo": -0.3435,
          "hi": 0.0686,
          "p": 0.1911
        },
        "n_rt": 949,
        "alpha_ts": {
          "alpha": 1.017,
          "lo": -1.0663,
          "hi": 3.1002,
          "t": 0.9568
        },
        "n_is": 1189,
        "alpha_f": {
          "alpha": 0.6667,
          "lo": -1.3751,
          "hi": 2.7084,
          "t": 0.64,
          "p": 0.5222
        }
      },
      "inv_vol": {
        "dsr": {
          "est": -0.0671,
          "lo": -0.2125,
          "hi": 0.0782,
          "p": 0.3651
        },
        "n_rt": 949,
        "alpha_ts": {
          "alpha": 0.8325,
          "lo": -0.7734,
          "hi": 2.4384,
          "t": 1.016
        },
        "n_is": 1189,
        "alpha_f": {
          "alpha": 0.6313,
          "lo": -0.9416,
          "hi": 2.2043,
          "t": 0.7867,
          "p": 0.4315
        }
      },
      "cap2": {
        "dsr": {
          "est": -0.1039,
          "lo": -0.2974,
          "hi": 0.0897,
          "p": 0.2928
        },
        "n_rt": 949,
        "alpha_ts": {
          "alpha": 0.5868,
          "lo": -0.755,
          "hi": 1.9286,
          "t": 0.8571
        },
        "n_is": 1189,
        "alpha_f": {
          "alpha": 0.4905,
          "lo": -0.8572,
          "hi": 1.8381,
          "t": 0.7134,
          "p": 0.4756
        }
      },
      "1947_1986": {
        "dsr": {
          "est": -0.3972,
          "lo": -0.6559,
          "hi": -0.1385,
          "p": 0.0026
        },
        "n_rt": 474,
        "alpha_ts": {
          "alpha": -1.8823,
          "lo": -4.8287,
          "hi": 1.064,
          "t": -1.2522
        },
        "n_is": 474,
        "alpha_f": {
          "alpha": -2.0562,
          "lo": -4.9057,
          "hi": 0.7933,
          "t": -1.4143,
          "p": 0.1573
        }
      },
      "1987_2026": {
        "dsr": {
          "est": -0.1517,
          "lo": -0.5266,
          "hi": 0.2231,
          "p": 0.4276
        },
        "n_rt": 475,
        "alpha_ts": {
          "alpha": -0.6548,
          "lo": -4.3695,
          "hi": 3.0599,
          "t": -0.3455
        },
        "n_is": 475,
        "alpha_f": {
          "alpha": -0.7394,
          "lo": -4.4824,
          "hi": 3.0036,
          "t": -0.3872,
          "p": 0.6986
        }
      },
      "2016_2026": {
        "dsr": {
          "est": -0.045,
          "lo": -0.6706,
          "hi": 0.5805,
          "p": 0.8878
        },
        "n_rt": 127,
        "alpha_ts": {
          "alpha": -0.0175,
          "lo": -4.3997,
          "hi": 4.3647,
          "t": -0.0078
        },
        "n_is": 127,
        "alpha_f": {
          "alpha": -0.26,
          "lo": -4.9747,
          "hi": 4.4546,
          "t": -0.1081,
          "p": 0.9139
        }
      }
    }
  },
  "verdict": {
    "Mkt-RF": {
      "H1": false,
      "H2": false,
      "H2_refuted": false,
      "H3a": false,
      "H3b": false,
      "survives": false
    },
    "SMB": {
      "H1": false,
      "H2": false,
      "H2_refuted": false,
      "H3a": false,
      "H3b": false,
      "survives": false
    },
    "HML": {
      "H1": false,
      "H2": false,
      "H2_refuted": false,
      "H3a": false,
      "H3b": false,
      "survives": false
    }
  },
  "forecast": {
    "Mkt-RF": {
      "rv3": {
        "sq": {
          "b": 0.4549,
          "t": 4.9859,
          "t_b1": -5.9755,
          "lo": 0.2761,
          "hi": 0.6337,
          "a": 15.6917,
          "r2": 0.1039,
          "n": 1189
        },
        "abs": {
          "b": 0.4312,
          "t": 7.163,
          "t_b1": -9.4473,
          "lo": 0.3132,
          "hi": 0.5492,
          "a": 1.9432,
          "r2": 0.121,
          "n": 1189
        }
      },
      "rv6": {
        "sq": {
          "b": 0.4836,
          "t": 4.6688,
          "t_b1": -4.9863,
          "lo": 0.2806,
          "hi": 0.6866,
          "a": 14.8877,
          "r2": 0.0798,
          "n": 1189
        },
        "abs": {
          "b": 0.4788,
          "t": 8.4677,
          "t_b1": -9.2158,
          "lo": 0.368,
          "hi": 0.5897,
          "a": 1.6481,
          "r2": 0.1224,
          "n": 1189
        }
      },
      "rv12": {
        "sq": {
          "b": 0.7547,
          "t": 4.3156,
          "t_b1": -1.4025,
          "lo": 0.412,
          "hi": 1.0975,
          "a": 7.1024,
          "r2": 0.147,
          "n": 1189
        },
        "abs": {
          "b": 0.5684,
          "t": 6.7116,
          "t_b1": -5.0965,
          "lo": 0.4024,
          "hi": 0.7344,
          "a": 1.1706,
          "r2": 0.1475,
          "n": 1189
        }
      }
    },
    "SMB": {
      "rv3": {
        "sq": {
          "b": 0.161,
          "t": 1.6843,
          "t_b1": -8.7742,
          "lo": -0.0264,
          "hi": 0.3485,
          "a": 8.3954,
          "r2": 0.0101,
          "n": 1189
        },
        "abs": {
          "b": 0.2804,
          "t": 3.7754,
          "t_b1": -9.6893,
          "lo": 0.1348,
          "hi": 0.426,
          "a": 1.4784,
          "r2": 0.0521,
          "n": 1189
        }
      },
      "rv6": {
        "sq": {
          "b": 0.2209,
          "t": 2.0927,
          "t_b1": -7.3822,
          "lo": 0.014,
          "hi": 0.4277,
          "a": 7.7975,
          "r2": 0.0106,
          "n": 1189
        },
        "abs": {
          "b": 0.3135,
          "t": 4.0166,
          "t_b1": -8.7955,
          "lo": 0.1605,
          "hi": 0.4665,
          "a": 1.3528,
          "r2": 0.0522,
          "n": 1189
        }
      },
      "rv12": {
        "sq": {
          "b": 0.3437,
          "t": 3.2126,
          "t_b1": -6.1358,
          "lo": 0.134,
          "hi": 0.5533,
          "a": 6.5711,
          "r2": 0.0155,
          "n": 1189
        },
        "abs": {
          "b": 0.3657,
          "t": 4.8899,
          "t_b1": -8.4804,
          "lo": 0.2191,
          "hi": 0.5123,
          "a": 1.1798,
          "r2": 0.0584,
          "n": 1189
        }
      }
    },
    "HML": {
      "rv3": {
        "sq": {
          "b": 0.3893,
          "t": 4.4628,
          "t_b1": -7.0012,
          "lo": 0.2183,
          "hi": 0.5603,
          "a": 7.806,
          "r2": 0.0838,
          "n": 1189
        },
        "abs": {
          "b": 0.4579,
          "t": 7.9017,
          "t_b1": -9.3556,
          "lo": 0.3443,
          "hi": 0.5715,
          "a": 1.0905,
          "r2": 0.1546,
          "n": 1189
        }
      },
      "rv6": {
        "sq": {
          "b": 0.4438,
          "t": 3.9494,
          "t_b1": -4.9502,
          "lo": 0.2235,
          "hi": 0.664,
          "a": 7.1077,
          "r2": 0.068,
          "n": 1189
        },
        "abs": {
          "b": 0.5225,
          "t": 9.3349,
          "t_b1": -8.5324,
          "lo": 0.4128,
          "hi": 0.6322,
          "a": 0.8397,
          "r2": 0.1715,
          "n": 1189
        }
      },
      "rv12": {
        "sq": {
          "b": 0.5858,
          "t": 3.9765,
          "t_b1": -2.812,
          "lo": 0.2971,
          "hi": 0.8745,
          "a": 5.2953,
          "r2": 0.0814,
          "n": 1189
        },
        "abs": {
          "b": 0.5575,
          "t": 7.0688,
          "t_b1": -5.6118,
          "lo": 0.4029,
          "hi": 0.712,
          "a": 0.6857,
          "r2": 0.1711,
          "n": 1189
        }
      }
    }
  },
  "risk_return": {
    "Mkt-RF": {
      "rv3": {
        "b": 0.3185,
        "t": 1.4261,
        "t_b1": -3.0518,
        "lo": -0.1192,
        "hi": 0.7562,
        "a": 0.688,
        "r2": 0.0036,
        "n": 1189
      },
      "rv6": {
        "b": 0.1058,
        "t": 0.567,
        "t_b1": -4.7899,
        "lo": -0.26,
        "hi": 0.4717,
        "a": 0.688,
        "r2": 0.0004,
        "n": 1189
      },
      "rv12": {
        "b": 0.3188,
        "t": 1.1288,
        "t_b1": -2.412,
        "lo": -0.2347,
        "hi": 0.8723,
        "a": 0.688,
        "r2": 0.0036,
        "n": 1189
      }
    },
    "SMB": {
      "rv3": {
        "b": -0.0414,
        "t": -0.3014,
        "t_b1": -7.5739,
        "lo": -0.3109,
        "hi": 0.2281,
        "a": 0.1722,
        "r2": 0.0002,
        "n": 1189
      },
      "rv6": {
        "b": -0.077,
        "t": -0.8394,
        "t_b1": -11.7477,
        "lo": -0.2566,
        "hi": 0.1027,
        "a": 0.1722,
        "r2": 0.0006,
        "n": 1189
      },
      "rv12": {
        "b": 0.1293,
        "t": 1.1658,
        "t_b1": -7.8474,
        "lo": -0.0881,
        "hi": 0.3468,
        "a": 0.1722,
        "r2": 0.0017,
        "n": 1189
      }
    },
    "HML": {
      "rv3": {
        "b": -0.0687,
        "t": -0.41,
        "t_b1": -6.3761,
        "lo": -0.3972,
        "hi": 0.2598,
        "a": 0.3489,
        "r2": 0.0004,
        "n": 1189
      },
      "rv6": {
        "b": -0.134,
        "t": -0.9174,
        "t_b1": -7.7628,
        "lo": -0.4203,
        "hi": 0.1523,
        "a": 0.3489,
        "r2": 0.0014,
        "n": 1189
      },
      "rv12": {
        "b": 0.2417,
        "t": 1.2825,
        "t_b1": -4.0231,
        "lo": -0.1277,
        "hi": 0.6111,
        "a": 0.3489,
        "r2": 0.0046,
        "n": 1189
      }
    }
  },
  "gap": {
    "Mkt-RF": {
      "vm_rt-f": {
        "est": -1.694,
        "lo": -4.3774,
        "hi": 0.9894,
        "p": 0.216
      },
      "vm_cap-f": {
        "est": -1.3067,
        "lo": -3.4553,
        "hi": 0.8419,
        "p": 0.2333
      },
      "vm_expost-f": {
        "est": -2.0009,
        "lo": -4.683,
        "hi": 0.6813,
        "p": 0.1437
      },
      "ts-f": {
        "est": -3.2704,
        "lo": -7.6155,
        "hi": 1.0746,
        "p": 0.1401
      },
      "c_fvm-f": {
        "est": -0.6626,
        "lo": -2.2267,
        "hi": 0.9015,
        "p": 0.4064
      },
      "c_fts-f": {
        "est": 0.4992,
        "lo": -1.844,
        "hi": 2.8424,
        "p": 0.6763
      },
      "c_all-f": {
        "est": 0.0096,
        "lo": -2.3567,
        "hi": 2.3759,
        "p": 0.9937
      }
    },
    "SMB": {
      "vm_rt-f": {
        "est": -1.8066,
        "lo": -3.9941,
        "hi": 0.3809,
        "p": 0.1055
      },
      "vm_cap-f": {
        "est": -1.6847,
        "lo": -3.2747,
        "hi": -0.0948,
        "p": 0.0378
      },
      "vm_expost-f": {
        "est": -1.6257,
        "lo": -3.7012,
        "hi": 0.4498,
        "p": 0.1247
      },
      "ts-f": {
        "est": 1.3857,
        "lo": -1.3749,
        "hi": 4.1463,
        "p": 0.3252
      },
      "c_fvm-f": {
        "est": -0.0978,
        "lo": -1.9271,
        "hi": 1.7315,
        "p": 0.9165
      },
      "c_fts-f": {
        "est": 1.2305,
        "lo": -0.9602,
        "hi": 3.4211,
        "p": 0.2709
      },
      "c_all-f": {
        "est": 0.9042,
        "lo": -1.5598,
        "hi": 3.3681,
        "p": 0.472
      }
    },
    "HML": {
      "vm_rt-f": {
        "est": -2.1481,
        "lo": -4.4455,
        "hi": 0.1492,
        "p": 0.0668
      },
      "vm_cap-f": {
        "est": -1.0145,
        "lo": -2.9229,
        "hi": 0.894,
        "p": 0.2975
      },
      "vm_expost-f": {
        "est": -2.2472,
        "lo": -4.523,
        "hi": 0.0286,
        "p": 0.0529
      },
      "ts-f": {
        "est": -0.7087,
        "lo": -4.0003,
        "hi": 2.5829,
        "p": 0.673
      },
      "c_fvm-f": {
        "est": -0.6965,
        "lo": -1.6765,
        "hi": 0.2836,
        "p": 0.1637
      },
      "c_fts-f": {
        "est": 0.6513,
        "lo": -0.9702,
        "hi": 2.2729,
        "p": 0.4311
      },
      "c_all-f": {
        "est": 0.0406,
        "lo": -1.6682,
        "hi": 1.7494,
        "p": 0.9629
      }
    }
  },
  "insample_dsr": {
    "Mkt-RF": {
      "vm-f": {
        "est": -0.066,
        "lo": -0.2632,
        "hi": 0.1312,
        "p": 0.5119
      },
      "ts-f": {
        "est": -0.1302,
        "lo": -0.4114,
        "hi": 0.1511,
        "p": 0.3644
      }
    },
    "SMB": {
      "vm-f": {
        "est": -0.1861,
        "lo": -0.3628,
        "hi": -0.0095,
        "p": 0.0388
      },
      "ts-f": {
        "est": 0.1092,
        "lo": -0.1438,
        "hi": 0.3622,
        "p": 0.3976
      }
    },
    "HML": {
      "vm-f": {
        "est": -0.174,
        "lo": -0.3699,
        "hi": 0.0219,
        "p": 0.0817
      },
      "ts-f": {
        "est": -0.0379,
        "lo": -0.3047,
        "hi": 0.2288,
        "p": 0.7805
      }
    }
  },
  "state_sr": {
    "Mkt-RF": {
      "high-low_var": {
        "est": -0.0131,
        "lo": -0.4602,
        "hi": 0.434,
        "p": 0.9543
      },
      "pos-neg_mom": {
        "est": 0.5068,
        "lo": 0.0746,
        "hi": 0.9389,
        "p": 0.0215
      }
    },
    "SMB": {
      "high-low_var": {
        "est": 0.1998,
        "lo": -0.1846,
        "hi": 0.5842,
        "p": 0.3082
      },
      "pos-neg_mom": {
        "est": 0.5547,
        "lo": 0.1557,
        "hi": 0.9537,
        "p": 0.0064
      }
    },
    "HML": {
      "high-low_var": {
        "est": 0.0485,
        "lo": -0.3476,
        "hi": 0.4447,
        "p": 0.8102
      },
      "pos-neg_mom": {
        "est": 0.5061,
        "lo": 0.1117,
        "hi": 0.9006,
        "p": 0.0119
      }
    }
  },
  "combo_capped": {
    "Mkt-RF": {
      "sr": {
        "c_fvm_cap": {
          "est": 0.5244,
          "lo": 0.269,
          "hi": 0.7798,
          "p": 0.0001
        },
        "c_fts_cap": {
          "est": 0.5908,
          "lo": 0.3666,
          "hi": 0.8151,
          "p": 0
        },
        "c_all_cap": {
          "est": 0.5631,
          "lo": 0.3422,
          "hi": 0.7839,
          "p": 0
        }
      },
      "dsr": {
        "c_fvm_cap-f": {
          "est": -0.033,
          "lo": -0.1038,
          "hi": 0.0378,
          "p": 0.3607
        },
        "c_fts_cap-f": {
          "est": 0.0335,
          "lo": -0.1231,
          "hi": 0.1901,
          "p": 0.6751
        },
        "c_all_cap-f": {
          "est": 0.0057,
          "lo": -0.1502,
          "hi": 0.1616,
          "p": 0.9427
        },
        "c_all_cap-c_fts_cap": {
          "est": -0.0278,
          "lo": -0.0618,
          "hi": 0.0063,
          "p": 0.1096
        },
        "c_all_cap-c_fvm_cap": {
          "est": 0.0387,
          "lo": -0.0999,
          "hi": 0.1774,
          "p": 0.5842
        }
      },
      "share_capped": {
        "c_fvm_cap": 0.0959,
        "c_fts_cap": 0,
        "c_all_cap": 0.0306
      },
      "impl": {
        "c_fvm_cap": {
          "mean_w": 1.2384,
          "max_w": 2,
          "share_w_gt2": 0,
          "turnover": 0.1004,
          "breakeven_bp": null
        },
        "c_fts_cap": {
          "mean_w": 1.1678,
          "max_w": 1.7162,
          "share_w_gt2": 0,
          "turnover": 0.1083,
          "breakeven_bp": 46.4454
        },
        "c_all_cap": {
          "mean_w": 1.1482,
          "max_w": 2,
          "share_w_gt2": 0,
          "turnover": 0.1545,
          "breakeven_bp": 5.5666
        }
      }
    },
    "SMB": {
      "sr": {
        "c_fvm_cap": {
          "est": 0.1587,
          "lo": -0.0697,
          "hi": 0.3872,
          "p": 0.1732
        },
        "c_fts_cap": {
          "est": 0.2441,
          "lo": 0.0155,
          "hi": 0.4726,
          "p": 0.0363
        },
        "c_all_cap": {
          "est": 0.2497,
          "lo": 0.0254,
          "hi": 0.4741,
          "p": 0.0291
        }
      },
      "dsr": {
        "c_fvm_cap-f": {
          "est": 0.0421,
          "lo": -0.0781,
          "hi": 0.1622,
          "p": 0.4926
        },
        "c_fts_cap-f": {
          "est": 0.1274,
          "lo": -0.1034,
          "hi": 0.3581,
          "p": 0.2792
        },
        "c_all_cap-f": {
          "est": 0.133,
          "lo": -0.1071,
          "hi": 0.3732,
          "p": 0.2776
        },
        "c_all_cap-c_fts_cap": {
          "est": 0.0057,
          "lo": -0.0501,
          "hi": 0.0614,
          "p": 0.8425
        },
        "c_all_cap-c_fvm_cap": {
          "est": 0.091,
          "lo": -0.1179,
          "hi": 0.2999,
          "p": 0.3933
        }
      },
      "share_capped": {
        "c_fvm_cap": 0.0211,
        "c_fts_cap": 0,
        "c_all_cap": 0.0211
      },
      "impl": {
        "c_fvm_cap": {
          "mean_w": 0.572,
          "max_w": 2,
          "share_w_gt2": 0,
          "turnover": 0.1647,
          "breakeven_bp": 19.3954
        },
        "c_fts_cap": {
          "mean_w": 0.3367,
          "max_w": 1.2302,
          "share_w_gt2": 0,
          "turnover": 0.1882,
          "breakeven_bp": 52.564
        },
        "c_all_cap": {
          "mean_w": 0.1921,
          "max_w": 2,
          "share_w_gt2": 0,
          "turnover": 0.2433,
          "breakeven_bp": 42.7676
        }
      }
    },
    "HML": {
      "sr": {
        "c_fvm_cap": {
          "est": 0.3487,
          "lo": 0.0888,
          "hi": 0.6086,
          "p": 0.0086
        },
        "c_fts_cap": {
          "est": 0.4446,
          "lo": 0.2145,
          "hi": 0.6747,
          "p": 0.0002
        },
        "c_all_cap": {
          "est": 0.4302,
          "lo": 0.1947,
          "hi": 0.6658,
          "p": 0.0003
        }
      },
      "dsr": {
        "c_fvm_cap-f": {
          "est": -0.0292,
          "lo": -0.0815,
          "hi": 0.023,
          "p": 0.2731
        },
        "c_fts_cap-f": {
          "est": 0.0667,
          "lo": -0.0993,
          "hi": 0.2327,
          "p": 0.4309
        },
        "c_all_cap-f": {
          "est": 0.0523,
          "lo": -0.1073,
          "hi": 0.2119,
          "p": 0.5204
        },
        "c_all_cap-c_fts_cap": {
          "est": -0.0144,
          "lo": -0.0651,
          "hi": 0.0363,
          "p": 0.5782
        },
        "c_all_cap-c_fvm_cap": {
          "est": 0.0815,
          "lo": -0.0622,
          "hi": 0.2253,
          "p": 0.2662
        }
      },
      "share_capped": {
        "c_fvm_cap": 0.0843,
        "c_fts_cap": 0,
        "c_all_cap": 0.0738
      },
      "impl": {
        "c_fvm_cap": {
          "mean_w": 1.1732,
          "max_w": 2,
          "share_w_gt2": 0,
          "turnover": 0.0629,
          "breakeven_bp": null
        },
        "c_fts_cap": {
          "mean_w": 0.8614,
          "max_w": 1.3592,
          "share_w_gt2": 0,
          "turnover": 0.1145,
          "breakeven_bp": 47.3752
        },
        "c_all_cap": {
          "mean_w": 1.0288,
          "max_w": 2,
          "share_w_gt2": 0,
          "turnover": 0.152,
          "breakeven_bp": 30.2741
        }
      }
    }
  },
  "paths": {
    "Mkt-RF": {
      "vol_pa": {
        "f": 14.9037,
        "vm_rt": 22.416,
        "ts": 15.024
      },
      "blocks": [
        {
          "from": "1947-07",
          "to": "1956-12",
          "w_vm": 1.8158,
          "w_vm_max": 8.4187,
          "long": 0.886,
          "cum_f": 150.22,
          "cum_vm_rt": 268.7099,
          "cum_ts": 103.88,
          "cum_vm_rt_m": 178.6565,
          "cum_ts_m": 103.0478
        },
        {
          "from": "1957-01",
          "to": "1966-12",
          "w_vm": 2.5171,
          "w_vm_max": 9.5162,
          "long": 0.6417,
          "cum_f": 218.37,
          "cum_vm_rt": 398.0809,
          "cum_ts": 116.21,
          "cum_vm_rt_m": 264.6711,
          "cum_ts_m": 115.279
        },
        {
          "from": "1967-01",
          "to": "1976-12",
          "w_vm": 1.1332,
          "w_vm_max": 5.3121,
          "long": 0.5917,
          "cum_f": 237.23,
          "cum_vm_rt": 420.4128,
          "cum_ts": 170.19,
          "cum_vm_rt_m": 279.5189,
          "cum_ts_m": 168.8266
        },
        {
          "from": "1977-01",
          "to": "1986-12",
          "w_vm": 1.0639,
          "w_vm_max": 3.5846,
          "long": 0.6583,
          "cum_f": 299.88,
          "cum_vm_rt": 378.6681,
          "cum_ts": 171.64,
          "cum_vm_rt_m": 251.7642,
          "cum_ts_m": 170.265
        },
        {
          "from": "1987-01",
          "to": "1996-12",
          "w_vm": 1.5869,
          "w_vm_max": 5.566,
          "long": 0.775,
          "cum_f": 395.71,
          "cum_vm_rt": 502.4445,
          "cum_ts": 209.77,
          "cum_vm_rt_m": 334.0591,
          "cum_ts_m": 208.0895
        },
        {
          "from": "1997-01",
          "to": "2006-12",
          "w_vm": 1.1893,
          "w_vm_max": 5.0239,
          "long": 0.725,
          "cum_f": 455.84,
          "cum_vm_rt": 584.1604,
          "cum_ts": 290.22,
          "cum_vm_rt_m": 388.3893,
          "cum_ts_m": 287.895
        },
        {
          "from": "2007-01",
          "to": "2016-12",
          "w_vm": 1.2384,
          "w_vm_max": 4.117,
          "long": 0.7667,
          "cum_f": 532.71,
          "cum_vm_rt": 664.0004,
          "cum_ts": 361.73,
          "cum_vm_rt_m": 441.4724,
          "cum_ts_m": 358.8321
        },
        {
          "from": "2017-01",
          "to": "2026-07",
          "w_vm": 1.2354,
          "w_vm_max": 12.0962,
          "long": 0.8261,
          "cum_f": 656.91,
          "cum_vm_rt": 786.5323,
          "cum_ts": 401.49,
          "cum_vm_rt_m": 522.9398,
          "cum_ts_m": 398.2736
        }
      ]
    },
    "SMB": {
      "vol_pa": {
        "f": 9.66,
        "vm_rt": 12.528,
        "ts": 9.6384
      },
      "blocks": [
        {
          "from": "1947-07",
          "to": "1956-12",
          "w_vm": 2.7569,
          "w_vm_max": 9.8313,
          "long": 0.2105,
          "cum_f": -24.93,
          "cum_vm_rt": -80.1752,
          "cum_ts": 11.39,
          "cum_vm_rt_m": -61.8213,
          "cum_ts_m": 11.4156
        },
        {
          "from": "1957-01",
          "to": "1966-12",
          "w_vm": 2.6278,
          "w_vm_max": 18.4601,
          "long": 0.5667,
          "cum_f": -6.22,
          "cum_vm_rt": -49.1028,
          "cum_ts": 36.92,
          "cum_vm_rt_m": -37.862,
          "cum_ts_m": 37.003
        },
        {
          "from": "1967-01",
          "to": "1976-12",
          "w_vm": 0.4933,
          "w_vm_max": 3.1491,
          "long": 0.55,
          "cum_f": 15.88,
          "cum_vm_rt": -60.0229,
          "cum_ts": 119.14,
          "cum_vm_rt_m": -46.2823,
          "cum_ts_m": 119.4077
        },
        {
          "from": "1977-01",
          "to": "1986-12",
          "w_vm": 1.0847,
          "w_vm_max": 5.1794,
          "long": 0.7167,
          "cum_f": 83.34,
          "cum_vm_rt": -46.0374,
          "cum_ts": 208.88,
          "cum_vm_rt_m": -35.4984,
          "cum_ts_m": 209.3494
        },
        {
          "from": "1987-01",
          "to": "1996-12",
          "w_vm": 0.8224,
          "w_vm_max": 3.3645,
          "long": 0.3833,
          "cum_f": 68.57,
          "cum_vm_rt": -66.8908,
          "cum_ts": 219.05,
          "cum_vm_rt_m": -51.5779,
          "cum_ts_m": 219.5422
        },
        {
          "from": "1997-01",
          "to": "2006-12",
          "w_vm": 0.4341,
          "w_vm_max": 1.6736,
          "long": 0.6,
          "cum_f": 98.82,
          "cum_vm_rt": -49.1821,
          "cum_ts": 224.92,
          "cum_vm_rt_m": -37.9232,
          "cum_ts_m": 225.4254
        },
        {
          "from": "2007-01",
          "to": "2016-12",
          "w_vm": 1.1993,
          "w_vm_max": 5.1001,
          "long": 0.5167,
          "cum_f": 113.39,
          "cum_vm_rt": -57.2757,
          "cum_ts": 198.27,
          "cum_vm_rt_m": -44.164,
          "cum_ts_m": 198.7155
        },
        {
          "from": "2017-01",
          "to": "2026-07",
          "w_vm": 0.6884,
          "w_vm_max": 3.4314,
          "long": 0.3304,
          "cum_f": 89.14,
          "cum_vm_rt": -69.6824,
          "cum_ts": 198.28,
          "cum_vm_rt_m": -53.7305,
          "cum_ts_m": 198.7256
        }
      ]
    },
    "HML": {
      "vol_pa": {
        "f": 9.7651,
        "vm_rt": 15.8097,
        "ts": 9.7851
      },
      "blocks": [
        {
          "from": "1947-07",
          "to": "1956-12",
          "w_vm": 2.9695,
          "w_vm_max": 18.6652,
          "long": 0.6404,
          "cum_f": 43.15,
          "cum_vm_rt": 93.2851,
          "cum_ts": 21.41,
          "cum_vm_rt_m": 57.619,
          "cum_ts_m": 21.3662
        },
        {
          "from": "1957-01",
          "to": "1966-12",
          "w_vm": 2.1166,
          "w_vm_max": 37.9536,
          "long": 0.7083,
          "cum_f": 81.56,
          "cum_vm_rt": 152.3947,
          "cum_ts": 57.5,
          "cum_vm_rt_m": 94.1291,
          "cum_ts_m": 57.3825
        },
        {
          "from": "1967-01",
          "to": "1976-12",
          "w_vm": 0.8512,
          "w_vm_max": 5.4496,
          "long": 0.675,
          "cum_f": 153.76,
          "cum_vm_rt": 188.0301,
          "cum_ts": 103.7,
          "cum_vm_rt_m": 116.1398,
          "cum_ts_m": 103.4881
        },
        {
          "from": "1977-01",
          "to": "1986-12",
          "w_vm": 0.8872,
          "w_vm_max": 3.2155,
          "long": 0.675,
          "cum_f": 212.79,
          "cum_vm_rt": 183.106,
          "cum_ts": 137.39,
          "cum_vm_rt_m": 113.0984,
          "cum_ts_m": 137.1092
        },
        {
          "from": "1987-01",
          "to": "1996-12",
          "w_vm": 1.303,
          "w_vm_max": 10.6619,
          "long": 0.5917,
          "cum_f": 239.73,
          "cum_vm_rt": 153.5215,
          "cum_ts": 133.99,
          "cum_vm_rt_m": 94.825,
          "cum_ts_m": 133.7161
        },
        {
          "from": "1997-01",
          "to": "2006-12",
          "w_vm": 0.8991,
          "w_vm_max": 6.4016,
          "long": 0.75,
          "cum_f": 314.03,
          "cum_vm_rt": 226.1179,
          "cum_ts": 216.67,
          "cum_vm_rt_m": 139.6654,
          "cum_ts_m": 216.2272
        },
        {
          "from": "2007-01",
          "to": "2016-12",
          "w_vm": 1.126,
          "w_vm_max": 5.781,
          "long": 0.375,
          "cum_f": 304.28,
          "cum_vm_rt": 211.196,
          "cum_ts": 213.28,
          "cum_vm_rt_m": 130.4487,
          "cum_ts_m": 212.8441
        },
        {
          "from": "2017-01",
          "to": "2026-07",
          "w_vm": 0.4694,
          "w_vm_max": 6.3805,
          "long": 0.5043,
          "cum_f": 291.85,
          "cum_vm_rt": 197.4648,
          "cum_ts": 236.29,
          "cum_vm_rt_m": 121.9674,
          "cum_ts_m": 235.8071
        }
      ]
    }
  },
  "fcast_weight": {
    "Mkt-RF": {
      "sr": {
        "est": 0.5149,
        "lo": 0.2567,
        "hi": 0.773,
        "p": 0.0001
      },
      "dsr_f": {
        "est": -0.0425,
        "lo": -0.1058,
        "hi": 0.0209,
        "p": 0.1887
      },
      "dsr_vm": {
        "est": 0.0712,
        "lo": -0.0682,
        "hi": 0.2106,
        "p": 0.3169
      },
      "gap_f": {
        "est": -0.6332,
        "lo": -1.5916,
        "hi": 0.3253,
        "p": 0.1954
      },
      "impl": {
        "mean_w": 1.3469,
        "max_w": 1.8581,
        "share_w_gt2": 0,
        "turnover": 0.0751,
        "breakeven_bp": null
      },
      "vol_pa": 19.1743,
      "corr_w_vm": 0.7426,
      "b_first": 0.3998,
      "b_min": 0.3998,
      "b_max": 0.4867,
      "b_last": 0.4835,
      "a_first": 40.5774,
      "a_last": 14.9111
    },
    "SMB": {
      "sr": {
        "est": 0.123,
        "lo": -0.1259,
        "hi": 0.372,
        "p": 0.3327
      },
      "dsr_f": {
        "est": 0.0064,
        "lo": -0.0334,
        "hi": 0.0461,
        "p": 0.7542
      },
      "dsr_vm": {
        "est": 0.1934,
        "lo": -0.0287,
        "hi": 0.4155,
        "p": 0.0879
      },
      "gap_f": {
        "est": 0.0614,
        "lo": -0.3287,
        "hi": 0.4514,
        "p": 0.7578
      },
      "impl": {
        "mean_w": 1.0709,
        "max_w": 1.2925,
        "share_w_gt2": 0,
        "turnover": 0.0224,
        "breakeven_bp": 23.1493
      },
      "vol_pa": 9.7633,
      "corr_w_vm": 0.2949,
      "b_first": 0.0331,
      "b_min": 0.0331,
      "b_max": 0.2473,
      "b_last": 0.2208,
      "a_first": 18.1641,
      "a_last": 7.802
    },
    "HML": {
      "sr": {
        "est": 0.3632,
        "lo": 0.099,
        "hi": 0.6274,
        "p": 0.007
      },
      "dsr_f": {
        "est": -0.0147,
        "lo": -0.1007,
        "hi": 0.0713,
        "p": 0.7375
      },
      "dsr_vm": {
        "est": 0.2053,
        "lo": 0.0111,
        "hi": 0.3994,
        "p": 0.0382
      },
      "gap_f": {
        "est": -0.1435,
        "lo": -0.9865,
        "hi": 0.6994,
        "p": 0.7386
      },
      "impl": {
        "mean_w": 1.4259,
        "max_w": 1.8617,
        "share_w_gt2": 0,
        "turnover": 0.0673,
        "breakeven_bp": null
      },
      "vol_pa": 12.3464,
      "corr_w_vm": 0.4725,
      "b_first": 0.3834,
      "b_min": 0.3834,
      "b_max": 0.4479,
      "b_last": 0.4438,
      "a_first": 19.3969,
      "a_last": 7.1135
    }
  },
  "joint": {
    "vm_f": {
      "alpha_pa": [
        1.6438,
        -1.1666,
        0.1401
      ],
      "wald": 3.1296,
      "df": 3,
      "p": 0.3721
    },
    "vm_f_ts": {
      "alpha_pa": [
        -0.3647,
        -0.8084,
        0.1944
      ],
      "wald": 0.7807,
      "df": 3,
      "p": 0.8541
    },
    "ts_f": {
      "alpha_pa": [
        6.4219,
        2.7505,
        3.2334
      ],
      "wald": 24.9304,
      "df": 3,
      "p": 0
    },
    "ts_f_vm": {
      "alpha_pa": [
        5.5209,
        2.5365,
        3.2363
      ],
      "wald": 23.805,
      "df": 3,
      "p": 0
    }
  },
  "joint_robustness": {
    "baseline": {
      "vm_f": {
        "alpha_pa": [
          1.6438,
          -1.1666,
          0.1401
        ],
        "wald": 3.1296,
        "df": 3,
        "p": 0.3721
      },
      "vm_f_ts": {
        "alpha_pa": [
          -0.3647,
          -0.8084,
          0.1944
        ],
        "wald": 0.7807,
        "df": 3,
        "p": 0.8541
      }
    },
    "rv3": {
      "vm_f": {
        "alpha_pa": [
          1.3114,
          0.9635,
          -0.7521
        ],
        "wald": 1.8271,
        "df": 3,
        "p": 0.6091
      },
      "vm_f_ts": {
        "alpha_pa": [
          0.124,
          0.9399,
          -0.7678
        ],
        "wald": 1.52,
        "df": 3,
        "p": 0.6777
      }
    },
    "rv12": {
      "vm_f": {
        "alpha_pa": [
          3.3493,
          -0.7892,
          0.6667
        ],
        "wald": 6.1974,
        "df": 3,
        "p": 0.1024
      },
      "vm_f_ts": {
        "alpha_pa": [
          1.6417,
          -0.2929,
          1.017
        ],
        "wald": 2.4743,
        "df": 3,
        "p": 0.4799
      }
    },
    "inv_vol": {
      "vm_f": {
        "alpha_pa": [
          1.2459,
          -0.6749,
          0.6313
        ],
        "wald": 4.1113,
        "df": 3,
        "p": 0.2497
      },
      "vm_f_ts": {
        "alpha_pa": [
          -0.2717,
          -0.3916,
          0.8325
        ],
        "wald": 1.7131,
        "df": 3,
        "p": 0.634
      }
    },
    "cap2": {
      "vm_f": {
        "alpha_pa": [
          0.9588,
          -0.8348,
          0.4905
        ],
        "wald": 3.8986,
        "df": 3,
        "p": 0.2726
      },
      "vm_f_ts": {
        "alpha_pa": [
          -0.7385,
          -0.5802,
          0.5868
        ],
        "wald": 2.1802,
        "df": 3,
        "p": 0.5359
      }
    },
    "1947_1986": {
      "vm_f": {
        "alpha_pa": [
          -1.4234,
          -2.3849,
          -2.0562
        ],
        "wald": 3.9465,
        "df": 3,
        "p": 0.2673
      },
      "vm_f_ts": {
        "alpha_pa": [
          -2.2686,
          -1.6305,
          -1.8823
        ],
        "wald": 3.4204,
        "df": 3,
        "p": 0.3312
      }
    },
    "1987_2026": {
      "vm_f": {
        "alpha_pa": [
          1.6357,
          -0.7543,
          -0.7394
        ],
        "wald": 1.6441,
        "df": 3,
        "p": 0.6494
      },
      "vm_f_ts": {
        "alpha_pa": [
          0.4228,
          -0.8014,
          -0.6548
        ],
        "wald": 1.0419,
        "df": 3,
        "p": 0.7911
      }
    },
    "2016_2026": {
      "vm_f": {
        "alpha_pa": [
          1.268,
          0.7482,
          -0.26
        ],
        "wald": 0.4429,
        "df": 3,
        "p": 0.9312
      },
      "vm_f_ts": {
        "alpha_pa": [
          2.0447,
          0.7347,
          -0.0175
        ],
        "wald": 0.4186,
        "df": 3,
        "p": 0.9364
      }
    }
  },
  "multiplicity": {
    "robustness_dsr": {
      "n": 24,
      "min_p": 0.0026,
      "min_holm": 0.0624,
      "n_raw_5": 4,
      "n_holm_5": 0
    },
    "realtime_dsr": {
      "n": 27,
      "min_p": 0.0339,
      "min_holm": 0.9153,
      "n_raw_5": 2,
      "n_holm_5": 0
    },
    "robustness_h1": {
      "n": 24,
      "min_p": 0.0339,
      "min_holm": 0.8136,
      "n_raw_5": 1,
      "n_holm_5": 0
    }
  },
  "n_surviving": 0,
  "hypothesis_supported": false
}
Experiment script · experiment_v1.py
"""Volatility-managed factors versus time-series factor momentum, in sample and in real time.

Real data only: the Fama-French monthly factors (Kenneth R. French Data Library), loaded with dataset.load().
Signals are known at the end of month t and applied to the return of month t+1.
Revision: adds a direct test of the variance signal, return-unit magnitudes, capped real-time combinations,
decade weight/return paths, in-sample Sharpe-ratio differences, state Sharpe-ratio tests and Holm adjustments.
Revision 2: a real-time weight on the fitted variance forecast of equation (8) instead of RV_t, and the H1 alpha under
every robustness specification with HAC Wald joint tests across the three factors (the companion edition's code).
No new random draws. Every earlier statistic is computed exactly as before (same code path, same seed).
"""
import json

import numpy as np
import pandas as pd
import statsmodels.api as sm
from scipy.stats import chi2, norm

from dataset import load

FACT = ["Mkt-RF", "SMB", "HML"]
NW, WIN, CAP, B, BLOCK, SEED = 12, 6, 2.0, 2000, 12, 20261024
IS0, OOS0, END = "1927-07", "1947-07", "2026-07"
SUBS = {"1947_1986": ("1947-07", "1986-12"), "1987_2026": ("1987-01", "2026-07"), "2016_2026": ("2016-01", "2026-07")}
Q = np.sqrt(12.0)


def r4(x):
    return None if x is None or not np.isfinite(x) else round(float(x), 4)


# ------------------------------------------------------------------ data: inspect, filter, report
df = load()
print("loaded", df.shape, {c: str(t) for c, t in df.dtypes.items()})
df = df.assign(Date=pd.to_datetime(df["Date"])).set_index("Date").sort_index()
n_raw, n_missing = len(df), int(df[FACT].isna().any(axis=1).sum())
df = df.dropna(subset=FACT)
gaps = int((np.diff(df.index.year * 12 + df.index.month) != 1).sum())
print(f"rows {n_raw}; rows with a missing factor value {n_missing}; calendar gaps {gaps}; {df.index[0]:%Y-%m} to {df.index[-1]:%Y-%m}")
print(df[FACT].describe().round(3).to_string())


# ------------------------------------------------------------------ inference
def hac(Y, L=NW):
    """Newey-West (Bartlett) long-run covariance of the rows of Y."""
    Y = Y - Y.mean(0)
    S = Y.T @ Y / len(Y)
    for lag in range(1, L + 1):
        G = Y[lag:].T @ Y[:-lag] / len(Y)
        S += (1 - lag / (L + 1)) * (G + G.T)
    return S


def sharpe(x):
    x = np.asarray(x, float)
    return Q * x.mean() / x.std()


def sr_test(a, b=None):
    """Annualised Sharpe ratio of a, or SR(a) - SR(b), with a Ledoit-Wolf (2008) HAC delta-method 95% interval."""
    X = np.column_stack([a] if b is None else [a, b]).astype(float)
    mu, g = X.mean(0), (X ** 2).mean(0)
    v = g - mu ** 2
    sign = np.array([1.0] if b is None else [1.0, -1.0])
    grad = np.concatenate([sign * g / v ** 1.5, -sign * mu / (2 * v ** 1.5)])
    se = np.sqrt(grad @ hac(np.column_stack([X, X ** 2])) @ grad / len(X))
    est = float(np.sum(sign * mu / np.sqrt(v)))
    return {"est": r4(Q * est), "lo": r4(Q * (est - 1.96 * se)), "hi": r4(Q * (est + 1.96 * se)), "p": r4(2 * norm.sf(abs(est / se)))}


def span(y, X):
    """OLS with Newey-West (12-lag) errors; intercept reported in percent per year."""
    m = sm.OLS(np.asarray(y, float), sm.add_constant(np.asarray(X, float))).fit(cov_type="HAC", cov_kwds={"maxlags": NW})
    a, se = m.params[0], m.bse[0]
    return {"alpha": r4(12 * a), "t": r4(m.tvalues[0]), "lo": r4(12 * (a - 1.96 * se)), "hi": r4(12 * (a + 1.96 * se)),
            "p": r4(m.pvalues[0]), "beta": [r4(b) for b in m.params[1:]], "beta_t": [r4(t) for t in m.tvalues[1:]],
            "r2_adj": r4(m.rsquared_adj), "n": int(m.nobs)}


def boot_dsr(a, b):
    """Circular block bootstrap (block 12) percentile interval for SR(a) - SR(b)."""
    rng = np.random.default_rng(SEED)
    T = len(a)
    starts = rng.integers(0, T, size=(B, int(np.ceil(T / BLOCK))))
    idx = ((starts[:, :, None] + np.arange(BLOCK)) % T).reshape(B, -1)[:, :T]
    A, Bb = np.asarray(a)[idx], np.asarray(b)[idx]
    d = Q * (A.mean(1) / A.std(1) - Bb.mean(1) / Bb.std(1))
    return [r4(x) for x in np.percentile(d, [2.5, 97.5])]


# ------------------------------------------------------------------ signals and strategies
def build(f, win=WIN, power=1.0):
    rv = (f ** 2).rolling(win).mean()                                  # realized variance, months t-win+1..t
    mom = np.expm1(np.log1p(f / 100).rolling(12).sum())                # compounded return, months t-11..t
    d = pd.DataFrame({"f": f, "wraw": 1 / rv.shift(1) ** power,
                      "s": np.sign(mom).replace(0, -1).shift(1),        # +1 if m_t > 0, else -1
                      "hv": (rv > rv.expanding().median()).astype(float).shift(1)}).loc[IS0:END]
    assert not d.isna().any().any()
    raw = d.f * d.wraw
    d["w_is"] = d.wraw * d.f.std() / raw.std()                         # full-sample c (Moreira-Muir)
    d["w_rt"] = d.wraw * (d.f.expanding().std() / raw.expanding().std()).shift(1)  # real-time c_t, data through t
    d["vm"], d["ts"] = d.w_is * d.f, d.s * d.f
    return d


def combo(d, cols, i0):
    """Expanding-window mean-variance combination of f, f/RV and s*f, rescaled to the factor's estimated variance."""
    X = np.column_stack([np.ones(len(d))] + [d[c].to_numpy() for c in cols])
    R = X * d.f.to_numpy()[:, None]
    n = np.arange(1, len(d) + 1)[:, None, None]
    mu = np.cumsum(R, 0) / n[:, :, 0]
    cov = (np.cumsum(R[:, :, None] * R[:, None, :], 0) - n * mu[:, :, None] * mu[:, None, :]) / np.maximum(n - 1, 1)
    idx = np.arange(i0 - 1, len(d) - 1)                                # estimates through month t ...
    th = np.linalg.solve(cov[idx], mu[idx][:, :, None])[:, :, 0]
    th *= np.sqrt(cov[idx, 0, 0] / np.einsum("ti,tij,tj->t", th, cov[idx], th))[:, None]
    return np.einsum("ti,ti->t", th, X[idx + 1])                       # ... set the weight on f for month t+1


def impl(w, f, base):
    """Leverage, turnover and the break-even one-way cost (basis points per unit of turnover)."""
    dw = np.abs(np.diff(w, prepend=w[0]))
    gap = lambda k: sharpe(w * f - k * dw) - base
    be = None
    if gap(0.0) > 0:
        lo, hi = 0.0, 50.0
        for _ in range(60):
            mid = (lo + hi) / 2
            lo, hi = (mid, hi) if gap(mid) > 0 else (lo, mid)
        be = 100 * lo
    return {"mean_w": r4(w.mean()), "max_w": r4(np.abs(w).max()), "share_w_gt2": r4(np.mean(np.abs(w) > 2)),
            "turnover": r4(dw[1:].mean()), "breakeven_bp": r4(be)}


def state(y, dummy):
    """Mean return by state and the Newey-West test of the difference (state=1 minus state=0)."""
    m = span(y, dummy)
    on, off = y[dummy == 1], y[dummy == 0]
    return {"mean1": r4(on.mean()), "mean0": r4(off.mean()), "sd1": r4(on.std()), "sd0": r4(off.std()),
            "sr1": r4(sharpe(on)), "sr0": r4(sharpe(off)), "diff": r4(m["beta"][0]), "diff_t": r4(m["beta_t"][0]),
            "diff_lo": r4(m["beta"][0] - 1.96 * m["beta"][0] / m["beta_t"][0]),
            "diff_hi": r4(m["beta"][0] + 1.96 * m["beta"][0] / m["beta_t"][0]), "n1": int(dummy.sum()), "n0": int((1 - dummy).sum())}


def variant(f, win=WIN, power=1.0, cap=None, period=None):
    """Real-time SR(VM) - SR(f) and the in-sample horse-race alpha for one specification or subperiod."""
    d = build(f, win, power)
    w_rt, w_is = (d.w_rt, d.w_is) if cap is None else (d.w_rt.clip(upper=cap), d.w_is.clip(upper=cap))
    p_rt, p_is = period or (OOS0, END), period or (IS0, END)
    o, s = d.loc[p_rt[0]:p_rt[1]], d.loc[p_is[0]:p_is[1]]
    vm = (w_is * d.f).loc[p_is[0]:p_is[1]]
    a, h1 = span(vm, s[["f", "ts"]]), span(vm, s[["f"]])
    return {"dsr": sr_test((w_rt.loc[o.index] * o.f).to_numpy(), o.f.to_numpy()), "n_rt": len(o),
            "alpha_ts": {k: a[k] for k in ("alpha", "lo", "hi", "t")}, "n_is": a["n"],
            "alpha_f": {k: h1[k] for k in ("alpha", "lo", "hi", "t", "p")}}      # Revision 2: H1 in every specification


# ------------------------------------------------------------------ revision helpers
def ols(y, x):
    """Slope of y on x (Newey-West, 12 lags): t for slope = 0 and for slope = 1, 95% interval, R-squared."""
    m = sm.OLS(np.asarray(y, float), sm.add_constant(np.asarray(x, float))).fit(cov_type="HAC", cov_kwds={"maxlags": NW})
    b, se = m.params[1], m.bse[1]
    return {"b": r4(b), "t": r4(b / se), "t_b1": r4((b - 1) / se), "lo": r4(b - 1.96 * se), "hi": r4(b + 1.96 * se),
            "a": r4(m.params[0]), "r2": r4(m.rsquared), "n": int(m.nobs)}


def delta(fn, X):
    """HAC (Bartlett, 12 lags) delta-method estimate and 95% interval of fn(means of [X, X**2])."""
    Y = np.column_stack([X, X ** 2]).astype(float)
    v = Y.mean(0)
    h = 1e-6 * np.maximum(np.abs(v), 1.0)
    g = np.array([(fn(v + e) - fn(v - e)) / (2 * e[i]) for i, e in enumerate(np.diag(h))])
    est, se = fn(v), np.sqrt(g @ hac(Y) @ g / len(Y))
    return {"est": r4(est), "lo": r4(est - 1.96 * se), "hi": r4(est + 1.96 * se), "p": r4(2 * norm.sf(abs(est / se)))}


def gap(a, f):
    """Mean-return gap (% per year) of strategy a over factor f after scaling a to f's volatility."""
    return delta(lambda v: 12 * (v[0] * np.sqrt(v[3] - v[1] ** 2) / np.sqrt(v[2] - v[0] ** 2) - v[1]), np.column_stack([a, f]))


def state_sr(f, h):
    """Annualised Sharpe ratio when h = 1 minus when h = 0, with a HAC delta-method interval."""
    def fn(v):  # E[fh], E[f(1-h)], E[h], E[f2 h], E[f2 (1-h)], E[h]
        m1, m0 = v[0] / v[2], v[1] / (1 - v[2])
        return Q * (m1 / np.sqrt(v[3] / v[2] - m1 ** 2) - m0 / np.sqrt(v[4] / (1 - v[2]) - m0 ** 2))
    return delta(fn, np.column_stack([f * h, f * (1 - h), h]))


def holm(ps):
    """Holm step-down adjusted p-values."""
    ps, run = np.asarray(ps, float), 0.0
    adj = np.empty(len(ps))
    for i, j in enumerate(np.argsort(ps)):
        run = max(run, min(1.0, (len(ps) - i) * ps[j]))
        adj[j] = run
    return adj


BLOCKS = [("1947-07", "1956-12")] + [(f"{y}-01", f"{y + 9}-12") for y in range(1957, 2017, 10)] + [("2017-01", "2026-07")]


# ------------------------------------------------------------------ revision 2 helpers
def fweight(d, i0):
    """Real-time weight c_t/(a_t + b_t RV_t). a_t, b_t: OLS of f_tau^2 on RV_tau-1 over tau <= t from 1927:07 (equation 8);
    c_t equates, over the same months, the volatility of f with that of f divided by the month-t fitted variance."""
    x, f = 1 / d.wraw.to_numpy(), d.f.to_numpy()
    idx = np.arange(i0 - 1, len(d) - 1)                                # estimates through month t ...
    n = idx + 1.0
    sx, sy, sxx, sxy = (np.cumsum(v)[idx] for v in (x, f ** 2, x * x, x * f ** 2))
    b = (sxy - sx * sy / n) / (sxx - sx ** 2 / n)
    a = (sy - b * sx) / n
    fit = a[:, None] + b[:, None] * x[None, :]                          # month-t fitted variance for every month
    past = np.arange(len(d))[None, :] <= idx[:, None]
    assert fit[past].min() > 0 and fit[np.arange(len(idx)), idx + 1].min() > 0
    c = np.nanstd(np.where(past, f, np.nan), 1, ddof=1) / np.nanstd(np.where(past, f / fit, np.nan), 1, ddof=1)
    return c / fit[np.arange(len(idx)), idx + 1], a, b                  # ... set the weight for month t+1


def joint(eqs):
    """HAC (Newey-West, 12 lags) Wald test that the intercepts of OLS equations on the same months are jointly zero."""
    psi, alpha, pos, k = [], [], [], 0
    for y, X in eqs:
        y, X = np.asarray(y, float), sm.add_constant(np.asarray(X, float))
        A = np.linalg.inv(X.T @ X / len(y))
        b = A @ X.T @ y / len(y)
        psi.append((X * (y - X @ b)[:, None]) @ A)
        alpha.append(b[0])
        pos.append(k)
        k += X.shape[1]
    V = hac(np.column_stack(psi))[np.ix_(pos, pos)] / len(y)
    a = np.array(alpha)
    w = float(a @ np.linalg.solve(V, a))
    return {"alpha_pa": [r4(12 * x) for x in a], "wald": r4(w), "df": len(a), "p": r4(chi2.sf(w, len(a)))}


def series(f, win=WIN, power=1.0, cap=None, period=None):
    """In-sample managed factor (full-sample c, optionally capped) and the signal frame for one specification."""
    d = build(f, win, power)
    w_is = d.w_is if cap is None else d.w_is.clip(upper=cap)
    p = period or (IS0, END)
    return (w_is * d.f).loc[p[0]:p[1]], d.loc[p[0]:p[1]]


# ------------------------------------------------------------------ estimation, factor by factor
out = {"sample": {"n_raw": n_raw, "n_missing": n_missing, "calendar_gaps": gaps, "first": f"{df.index[0]:%Y-%m}",
                  "last": f"{df.index[-1]:%Y-%m}", "insample": f"{IS0} to {END}", "oos": f"{OOS0} to {END}"},
       "settings": {"nw_lags": NW, "rv_window": WIN, "leverage_cap": CAP, "boot_draws": B, "boot_block": BLOCK, "seed": SEED},
       "summary": {}, "mechanism": {}, "spanning": {}, "realtime": {}, "implementation": {}, "robustness": {}, "verdict": {},
       "forecast": {}, "risk_return": {}, "gap": {}, "insample_dsr": {}, "state_sr": {}, "combo_capped": {}, "paths": {},
       "fcast_weight": {}}
D = {}
for k in FACT:
    f = df[k]
    d = build(f)
    D[k] = d
    i0 = d.index.get_loc(pd.Timestamp(OOS0))
    out["sample"]["n_insample"], out["sample"]["n_oos"] = len(d), len(d) - i0
    out["sample"]["n_2016_2026"] = len(d.loc["2016-01":END])

    # Table 1: summary statistics and signal behaviour, in-sample period
    out["summary"][k] = {"mean_pa": r4(12 * d.f.mean()), "sd_pa": r4(Q * d.f.std()), "sr": sr_test(d.f.to_numpy()),
                         "sr_vm": sr_test(d.vm.to_numpy()), "sr_ts": sr_test(d.ts.to_numpy()),
                         "share_long": r4((d.s > 0).mean()), "c_full": r4(d.f.std() / (d.f * d.wraw).std()),
                         "corr_w_s": r4(np.corrcoef(d.wraw, d.s)[0, 1]),
                         "w_is_long": r4(d.w_is[d.s > 0].mean()), "w_is_short": r4(d.w_is[d.s < 0].mean()),
                         "w_is_max": r4(d.w_is.max()), "share_w_is_gt2": r4((d.w_is > 2).mean())}
    out["mechanism"][k] = {"high_vs_low_var": state(d.f, d.hv), "pos_vs_neg_mom": state(d.f, (d.s > 0).astype(float))}

    # Table 2: spanning regressions and the in-sample horse race (Newey-West, 12 lags)
    out["spanning"][k] = {"vm_f": span(d.vm, d[["f"]]), "vm_f_ts": span(d.vm, d[["f", "ts"]]),
                          "ts_f": span(d.ts, d[["f"]]), "ts_f_vm": span(d.ts, d[["f", "vm"]])}

    # Table 3: real-time out-of-sample comparisons, 1947:07-2026:07
    o = d.iloc[i0:]
    fo = o.f.to_numpy()
    W = {"f": np.ones(len(o)), "vm_rt": o.w_rt.to_numpy(), "vm_cap": o.w_rt.clip(upper=CAP).to_numpy(),
         "vm_expost": o.w_is.to_numpy(), "ts": o.s.to_numpy(), "c_fvm": combo(d, ["wraw"], i0),
         "c_fts": combo(d, ["s"], i0), "c_all": combo(d, ["wraw", "s"], i0)}
    Rr = {n: w * fo for n, w in W.items()}
    pairs = {"vm_rt-f": ("vm_rt", "f"), "vm_cap-f": ("vm_cap", "f"), "vm_expost-f": ("vm_expost", "f"), "ts-f": ("ts", "f"),
             "c_fvm-f": ("c_fvm", "f"), "c_fts-f": ("c_fts", "f"), "c_all-f": ("c_all", "f"),
             "c_all-c_fts": ("c_all", "c_fts"), "c_all-c_fvm": ("c_all", "c_fvm")}
    out["realtime"][k] = {"sr": {n: sr_test(r) for n, r in Rr.items()},
                          "dsr": {n: sr_test(Rr[a], Rr[b]) for n, (a, b) in pairs.items()},
                          "boot_vm_rt-f": boot_dsr(Rr["vm_rt"], Rr["f"]), "n": len(o)}
    base = sharpe(fo)
    out["implementation"][k] = {n: impl(W[n], fo, base) for n in ("vm_rt", "vm_cap", "ts", "c_fvm", "c_fts", "c_all")}

    # Revision: does RV_t forecast next-month variance (f^2, |f|), and how does next-month return move with RV_t?
    rv = {w: 1 / build(f, w).wraw for w in (3, 6, 12)}                 # RV_t aligned with f_{t+1}, 1927:07-2026:07
    out["forecast"][k] = {f"rv{w}": {"sq": ols(d.f ** 2, rv[w]), "abs": ols(d.f.abs(), np.sqrt(rv[w]))} for w in rv}
    out["risk_return"][k] = {f"rv{w}": ols(d.f, (rv[w] - rv[w].mean()) / rv[w].std()) for w in rv}
    # Revision: return-unit magnitudes, in-sample Sharpe differences, tests of the state Sharpe ratios
    out["gap"][k] = {n: gap(Rr[a], Rr["f"]) for n, (a, b) in pairs.items() if b == "f"}
    out["insample_dsr"][k] = {"vm-f": sr_test(d.vm.to_numpy(), d.f.to_numpy()), "ts-f": sr_test(d.ts.to_numpy(), d.f.to_numpy())}
    out["state_sr"][k] = {"high-low_var": state_sr(d.f.to_numpy(), d.hv.to_numpy()),
                          "pos-neg_mom": state_sr(d.f.to_numpy(), (d.s > 0).astype(float).to_numpy())}
    # Revision: real-time combinations with the weight on the factor capped at |W| <= 2
    Wc = {n + "_cap": np.clip(W[n], -CAP, CAP) for n in ("c_fvm", "c_fts", "c_all")}
    Rc = {**{n: w * fo for n, w in Wc.items()}, "f": fo}
    cpairs = [("c_fvm_cap", "f"), ("c_fts_cap", "f"), ("c_all_cap", "f"), ("c_all_cap", "c_fts_cap"), ("c_all_cap", "c_fvm_cap")]
    out["combo_capped"][k] = {"sr": {n: sr_test(Rc[n]) for n in Wc}, "dsr": {f"{a}-{b}": sr_test(Rc[a], Rc[b]) for a, b in cpairs},
                              "share_capped": {n: r4(np.mean(np.abs(W[n[:-4]]) > CAP)) for n in Wc},
                              "impl": {n: impl(Wc[n], fo, base) for n in Wc}}
    # Revision: decade paths of the real-time weights and cumulative returns (sums of monthly % returns from 1947:07),
    # raw and with VM and TSFM rescaled ex post to the factor's 1947-2026 volatility ("_m"; descriptive only)
    cum = pd.DataFrame({**{n: Rr[n] for n in ("f", "vm_rt", "ts")},
                        **{n + "_m": Rr[n] * fo.std() / Rr[n].std() for n in ("vm_rt", "ts")}}, index=o.index).cumsum()
    out["paths"][k] = {"vol_pa": {n: r4(Q * Rr[n].std()) for n in ("f", "vm_rt", "ts")},
                       "blocks": [{"from": a, "to": b, "w_vm": r4(o.w_rt.loc[a:b].mean()), "w_vm_max": r4(o.w_rt.loc[a:b].max()),
                                   "long": r4((o.s.loc[a:b] > 0).mean()), **{f"cum_{n}": r4(cum[n].loc[:b].iloc[-1]) for n in cum}}
                                  for a, b in BLOCKS]}
    # Revision 2: weight on the real-time variance forecast a_t + b_t RV_t instead of RV_t (the noisy-signal test)
    w_fc, a_fc, b_fc = fweight(d, i0)
    r_fc = w_fc * fo
    out["fcast_weight"][k] = {"sr": sr_test(r_fc), "dsr_f": sr_test(r_fc, fo), "dsr_vm": sr_test(r_fc, Rr["vm_rt"]),
                              "gap_f": gap(r_fc, fo), "impl": impl(w_fc, fo, base), "vol_pa": r4(Q * r_fc.std()),
                              "corr_w_vm": r4(np.corrcoef(w_fc, W["vm_rt"])[0, 1]),
                              "b_first": r4(b_fc[0]), "b_min": r4(b_fc.min()), "b_max": r4(b_fc.max()), "b_last": r4(b_fc[-1]),
                              "a_first": r4(a_fc[0]), "a_last": r4(a_fc[-1])}
    print(f"{k}: forecast weight dSR vs f {out['fcast_weight'][k]['dsr_f']}; vs VM {out['fcast_weight'][k]['dsr_vm']}")

    # Table 4: robustness of the two headline statistics
    out["robustness"][k] = {"baseline": variant(f), "rv3": variant(f, win=3), "rv12": variant(f, win=12),
                            "inv_vol": variant(f, power=0.5), "cap2": variant(f, cap=CAP),
                            **{p: variant(f, period=SUBS[p]) for p in SUBS}}

    sp, dsr = out["spanning"][k], out["realtime"][k]["dsr"]
    v = {"H1": sp["vm_f"]["lo"] > 0, "H2": dsr["vm_rt-f"]["lo"] > 0, "H2_refuted": dsr["vm_rt-f"]["hi"] < 0,
         "H3a": sp["vm_f_ts"]["lo"] > 0, "H3b": dsr["c_all-c_fts"]["lo"] > 0}
    v["survives"] = v["H2"] and v["H3a"]
    out["verdict"][k] = v
    print(f"{k}: alpha(VM|f) {sp['vm_f']['alpha']} [{sp['vm_f']['lo']}, {sp['vm_f']['hi']}]; alpha(VM|f,TS) {sp['vm_f_ts']['alpha']}; "
          f"real-time dSR {dsr['vm_rt-f']['est']} [{dsr['vm_rt-f']['lo']}, {dsr['vm_rt-f']['hi']}]; verdict {v}")

# Revision 2: HAC Wald joint tests across the three factors, at baseline and under every robustness specification
assert all(D[k].index.equals(D[FACT[0]].index) for k in FACT)
out["joint"] = {"vm_f": joint([(D[k].vm, D[k][["f"]]) for k in FACT]), "vm_f_ts": joint([(D[k].vm, D[k][["f", "ts"]]) for k in FACT]),
                "ts_f": joint([(D[k].ts, D[k][["f"]]) for k in FACT]), "ts_f_vm": joint([(D[k].ts, D[k][["f", "vm"]]) for k in FACT])}
SPECS = {"baseline": {}, "rv3": {"win": 3}, "rv12": {"win": 12}, "inv_vol": {"power": 0.5}, "cap2": {"cap": CAP},
         **{p: {"period": SUBS[p]} for p in SUBS}}
out["joint_robustness"] = {}
for n, kw in SPECS.items():
    S = {k: series(df[k], **kw) for k in FACT}
    assert all(S[k][1].index.equals(S[FACT[0]][1].index) for k in FACT)
    out["joint_robustness"][n] = {"vm_f": joint([(S[k][0], S[k][1][["f"]]) for k in FACT]),
                                  "vm_f_ts": joint([(S[k][0], S[k][1][["f", "ts"]]) for k in FACT])}
print("joint HAC Wald p, H1 by specification:", {n: v["vm_f"]["p"] for n, v in out["joint_robustness"].items()})

pA = [out["robustness"][k][v]["dsr"]["p"] for k in FACT for v in out["robustness"][k]]
pB = [out["realtime"][k]["dsr"][n]["p"] for k in FACT for n in out["realtime"][k]["dsr"]]
pC = [out["robustness"][k][v]["alpha_f"]["p"] for k in FACT for v in out["robustness"][k]]
out["multiplicity"] = {nm: {"n": len(p), "min_p": r4(min(p)), "min_holm": r4(holm(p).min()), "n_raw_5": int(sum(x < 0.05 for x in p)),
                            "n_holm_5": int((holm(p) < 0.05).sum())}
                       for nm, p in (("robustness_dsr", pA), ("realtime_dsr", pB), ("robustness_h1", pC))}
out["n_surviving"] = int(sum(out["verdict"][k]["survives"] for k in FACT))
out["hypothesis_supported"] = bool(out["n_surviving"] >= 2)
print("RESULT_JSON: " + json.dumps(out, separators=(",", ":")))
Listing. experiment_v1.py

4.Analysis

Attempt 4Done

Analysis complete: experiment succeeded; hypothesis refuted; 18 key numbers.

Reviewer feedback applied: Independent referee (Claude Fable 5.1), round 3: accept, with proof corrections: (1) [text] Cut the abstract to 100 words or fewer (it is 117): for example drop 'across eight specifications' and the parenthetical '(3.35% a year)', and merge the last two sentences. (2) [text] In the abstract's last sentence and in the conclusion's 'accounts for most of that loss', add the level at which the forecast weight beats inverse-variance weighting (5% for value, 10% for size, not for the market), as Section IV.D already states. Keep the analysis record consistent with the paper; no other change.
Hypothesis refutedExperiment succeeded

Results and analysis

The extended analysis ran on the real Fama-French monthly factors (1,201 months, July 1926 to July 2026; 1,189 in-sample and 949 real-time months), reproduces every earlier statistic exactly and adds no random draws. It tests volatility management measurable from monthly returns (a six-month realized variance), not the daily-return signal of Moreira and Muir, which these data cannot build. Under the pre-committed rule VM survives for none of the three factors, so the hypothesis is not supported (refuted); only the narrower question of whether real-time VM is worse than its factor stays inconclusive, because no H2 interval lies entirely below zero. H1 fails in the baseline (market 1.64% a year, t = 0.90; HAC Wald joint p = 0.37). Across eight specifications, one of 24 H1 alphas, the market's with a twelve-month window, is significant at the 5% level (3.35% a year [0.26, 6.44]), but the three twelve-month alphas are not jointly significant (p = 0.10), no other cell is significant at the 10% level, and none survives a Holm adjustment (smallest adjusted p 0.81). H3a fails (-0.36%, t = -0.23; momentum loading 0.31, t = 5.64), while TSFM alphas of 5.52%, 2.54% and 3.24% survive the volatility control. H2 fails: real-time Sharpe differences are -0.11, -0.19 and -0.22, and gains above 0.07, 0.04 and 0.01 are rejected. The direct test of the noisy-signal explanation weights each factor by the inverse of its real-time variance forecast from the expanding-window regression of squared returns on RV_t. This weight beats 1/RV_t by 0.07, 0.19 and 0.21 (significant at the 5% level for HML and 10% for SMB, not for Mkt-RF) and removes most of the shortfall (-0.04, 0.01 and -0.01 against the factor, none significant at 10%), but it adds no gain: gains above 0.02, 0.05 and 0.07 are rejected. Noise in the signal thus accounts for most of VM's measured loss, significantly so for HML and SMB but not for Mkt-RF; the forecastable part of monthly variance carries no detectable timing value.

Key numbers

MetricValueInterpretation
Sample1,201 months (1926:07-2026:07), none dropped; in-sample N = 1,189; real-time N = 949; 2016-2026 N = 127All tests use the stated samples.
Scope of the signalsix-month realized variance of monthly returns (three- and twelve-month windows as robustness)Not the previous-month daily realized variance of Moreira and Muir (2017); the daily file is not in this study's data.
VM alpha on own factor (H1), baselineMkt-RF 1.64% p.a. (t = 0.90) [-1.92, 5.20]; SMB -1.17% (t = -1.25); HML 0.14% (t = 0.13); HAC Wald joint test p = 0.37No alpha at the 10% level, singly or jointly.
H1 alpha under the eight robustness specifications1 of 24 significant at 5%: Mkt-RF with a twelve-month window, 3.35% p.a. (t = 2.12) [0.26, 6.44]; joint p for that specification 0.10 (Wald 6.20, 3 df); joint p 0.10 to 0.93 across specifications; Holm-adjusted p of the 24 cells at least 0.81An isolated cell among 24 overlapping ones, not jointly significant in any specification; the same numbers as the companion Chinese edition.
VM alpha controlling for TSFM (H3a)Mkt-RF -0.36% (t = -0.23), TSFM loading 0.31 (t = 5.64); SMB -0.81%; HML 0.19%; joint p = 0.85 (0.33 to 0.94 across specifications)No horse-race alpha, singly or jointly; the twelve-month market alpha halves to 1.64% [-0.93, 4.21] with TSFM.
TSFM alpha controlling for VMMkt-RF 5.52% (t = 3.13); SMB 2.54% (t = 2.38); HML 3.24% (t = 2.87); joint Wald 23.81, p < 0.001Momentum alphas survive the volatility control.
Real-time SR(VM) - SR(f) (H2)Mkt-RF -0.11 [-0.29, 0.07]; SMB -0.19 [-0.41, 0.04]; HML -0.22 [-0.45, 0.01]Gains above 0.07, 0.04 and 0.01 in annualised Sharpe ratio are rejected; no interval lies entirely below zero.
Gap at matched volatility, 1947-2026Mkt-RF -1.69% p.a. [-4.38, 0.99]; SMB -1.81% [-3.99, 0.38]; HML -2.15% [-4.45, 0.15]The real-time shortfall in return units.
Weight on the real-time variance forecast: SR differences (noisy-signal test)against f: Mkt-RF -0.04 [-0.11, 0.02]; SMB 0.01 [-0.03, 0.05]; HML -0.01 [-0.10, 0.07]; against real-time VM: 0.07 [-0.07, 0.21]; 0.19 [-0.03, 0.42]; 0.21 [0.01, 0.40]Noise in RV_t accounts for most of VM's measured shortfall, significantly for HML (5% level) and SMB (10%) but not for Mkt-RF; the calibrated forecast adds no gain, and gains above 0.02, 0.05 and 0.07 are rejected.
Weight on the real-time variance forecast: magnitudes and tradinggap at matched volatility -0.63%, 0.06% and -0.14% p.a.; maximum weight 1.86, 1.29 and 1.86 (VM 12.10, 18.46, 37.95); turnover 0.08, 0.02 and 0.07 a month (VM 0.37, 0.36, 0.46)With forecast slopes well below one, the calibrated weight moves little and stays close to the factor.
In-sample SR(VM) - SR(f), 1927-2026Mkt-RF -0.07 [-0.26, 0.13]; SMB -0.19 [-0.36, -0.01]; HML -0.17 [-0.37, 0.02]Significant at 5% for SMB, 10% for HML.
Capped real-time combinations (|W| <= 2), H3bSR({f,VM,TSFM}) - SR({f,TSFM}): -0.03 [-0.06, 0.01]; 0.01 [-0.05, 0.06]; -0.01 [-0.07, 0.04]H3b does not hinge on extreme unregularised weights.
Variance forecast, f^2(t+1) on RV(t), six-month windowslopes 0.48 (t = 4.67), 0.22 (t = 2.09), 0.44 (t = 3.95); t for slope = 1: -4.99, -7.38, -4.95; R2 0.08, 0.01, 0.07Slopes far below one are consistent with a noisy signal; the twelve-month window forecasts best (slopes 0.75, 0.34, 0.59).
One-standard-deviation effect of RV(t) on next-month returnMkt-RF 0.11% per month (t = 0.57); SMB -0.08% (t = -0.84); HML -0.13% (t = -0.92), six-month windowNo detectable risk-return relation in either direction.
State Sharpe-ratio testshigh minus low variance: -0.01 [-0.46, 0.43], 0.20 [-0.18, 0.58], 0.05 [-0.35, 0.44]; positive minus negative prior year: 0.51 [0.07, 0.94], 0.55 [0.16, 0.95], 0.51 [0.11, 0.90]Timing information is in past returns, not in past variance.
Multiplicityreal-time robustness differences: 4 of 24 significant at 5% individually, 0 after Holm (smallest adjusted p 0.06); real-time differences: 2 of 27, 0 after Holm (0.92); H1 robustness alphas: 1 of 24, 0 after Holm (0.81)Isolated significant cells should not be read individually.
Real-time volatility, 1947-2026VM 22.42% vs factor 14.90% (Mkt-RF); 12.53% vs 9.66% (SMB); 15.81% vs 9.77% (HML)The real-time constant leaves the managed factors more volatile than the factors.
Decade paths (volatility-matched cumulative returns, 1947:07-2026:07)Mkt-RF: f 656.91, VM 522.94, TSFM 398.27; SMB: 89.14, -53.73, 198.73; HML: 291.85, 121.97, 235.81 (sums of monthly %)The managed factors' cumulative lead was built in 1947-1966, when their mean weights were 1.8 to 3.0; VM fell behind the market factor by 1986 and behind HML by 1976.

Caveats

  • The signal is a six-month realized variance of monthly returns. The daily-return signal of Moreira and Muir (2017) is not tested, and that test is the decisive next step.
  • Only three factors; momentum, profitability and investment factors are not in the data.
  • Real-time Sharpe differences have low power. The intervals bound the gains but do not establish losses at the 5% level (except SMB in sample and some individual robustness cells, none of which survives a Holm adjustment).
  • The one significant H1 cell (market, twelve-month window) is one of 24 overlapping cells that share months and data; it is not jointly significant with the other factors and does not survive a Holm adjustment.
  • The forecast-based weight uses one linear forecast of next-month squared returns on the six-month RV_t, estimated on an expanding window from 1927; richer forecasts (more lags, other windows, daily data) are not tested.
  • Returns are gross of costs, financing spreads and margin constraints; factor histories are the current vintage; pre-1963 coverage is thinner.
  • The forecast and risk-return regressions of Table IV Panel C are in sample (1927-2026); standardising RV_t by its full-sample moments only rescales the slope.
  • The decade paths are descriptive. The volatility-matched series are rescaled with the full 1947-2026 volatility, after the fact.

5.Paper writing

Attempt 10Done

Manuscript drafted: “Volatility Management from Monthly Returns versus Factor Momentum: The Fama-French Market, Size, and Value Factors, 1926–2026” (≈ 7965 words).

Reviewer feedback applied: Independent referee (Claude Fable 5.1), round 3: accept, with proof corrections: (1) [text] Cut the abstract to 100 words or fewer (it is 117): for example drop 'across eight specifications' and the parenthetical '(3.35% a year)', and merge the last two sentences. (2) [text] In the abstract's last sentence and in the conclusion's 'accounts for most of that loss', add the level at which the forecast weight beats inverse-variance weighting (5% for value, 10% for size, not for the market), as Section IV.D already states. Keep the analysis record consistent with the paper; no other change.

≈ 7965 words · AI-drafted manuscript · not peer-reviewed

Volatility Management from Monthly Returns versus Factor Momentum: The Fama-French Market, Size, and Value Factors, 1926–2026

Abstract

We test volatility management from monthly returns, not Moreira and Muir's daily signal, on the three Fama-French factors, 1926–2026. Its alphas are indistinguishable from zero with or without a momentum control, and momentum alphas survive a volatility control; of 24 benchmark alphas only the market's twelve-month-window alpha is significant at 5%, though not jointly with size and value (p = 0.10). Real-time annualized Sharpe ratios fall by 0.11 to 0.22 (gains above 0.07 rejected); a variance-forecast weight removes most of this loss, significantly for value (5% level) and size (10%) but not the market, without a detectable gain.

Keywords: volatility-managed portfolios; factor momentum; time-series momentum; real-time implementation; variance forecasting; Sharpe-ratio tests

I. Introduction

Moreira and Muir (2017) scale a factor by the inverse of its previous-month realized variance from daily returns and find large spanning alphas and higher Sharpe ratios for the market, value, momentum, profitability, and other factors, because volatility changes are not matched by proportional changes in expected returns; Moreira and Muir (2019) extend the question to long-horizon investors. The 2017 result builds on the value of volatility timing (Fleming, Kirby, and Ostdiek 2001), especially with high-frequency realized variance (Fleming, Kirby, and Ostdiek 2003; Bollerslev, Hood, Huss, and Pedersen 2018), sits uneasily with a positive risk-return trade-off (French, Schwert, and Stambaugh 1987), and fits slow-moving volatility expectations (Lochstoer and Muir 2022). Our question is narrower: can a variance measured from monthly returns alone time the three Fama-French factors?

Whether investors could have captured these gains is contested. Liu, Tang, and Zhou (2019) show that the usual market application sets the constant with full-sample information and, corrected, suffers deep drawdowns; Cederburg, O'Doherty, Wang, and Yan (2020) find across 103 strategies that managed portfolios do not systematically beat unmanaged ones and that spanning-implied combinations fail out of sample; Barroso and Detzel (2021) find that after costs only the managed market survives. DeMiguel, Martín-Utrera, and Uppal (2024) and Xu (2026) report real-time gains for multifactor and refined versions.

A second literature times factors, whose predictability lets timing improve on static factor investing (Haddad, Kozak, and Santosh 2020), and much of it uses their own past returns (Moskowitz, Ooi, and Pedersen 2012; Gupta and Kelly 2019; Ehsani and Linnainmaa 2022; Arnott, Kalesnik, and Linnainmaa 2023). The two signals are linked: volatility rises after bad news (Campbell and Hentschel 1992; Glosten, Jagannathan, and Runkle 1993), so an inverse-variance weight is lowest when a factor's trailing return is negative, which is when a momentum rule is short. Volatility scaling is best established for the momentum factor (Barroso and Santa-Clara 2015; Daniel and Moskowitz 2016), most of the alpha of time-series momentum comes from its volatility scaling (Kim, Tse, and Wald 2016), and asset-by-asset evidence for time-series momentum is weak (Huang, Li, Wang, and Zhou 2020).

We study the Fama and French (1993) market, size, and value factors from July 1926 to July 2026. Our signal is a trailing six-month realized variance of monthly returns, not the previous-month daily realized variance of Moreira and Muir (2017), which our monthly data cannot construct, so we do not test the published rule; a test with daily returns is the decisive next step. We pit this monthly volatility management (VM) against a sign-only time-series factor-momentum (TSFM) rule, run the horse race both ways in sample and in real time from July 1947, and test whether the signal forecasts variance and whether a weight on its real-time forecast does better. The hypotheses and decision rule were fixed before estimation; the diagnostics do not enter them.

Four results emerge. First, baseline VM alphas are indistinguishable from zero, singly and jointly, with or without a momentum control (market 1.64% a year, t = 0.90; −0.36%, t = −0.23). Of the 24 benchmark alphas (VM on its factor) across eight specifications, only the market's with a twelve-month window is significant at the 5% level (3.35% a year), but the three twelve-month alphas are not jointly significant (p = 0.10) and the cell does not survive a Holm adjustment. The in-sample content is the market's mechanical momentum loading of 0.31 (t = 5.64), and TSFM alphas of 5.52%, 2.54%, and 3.24% a year survive the volatility control. Second, in real time VM lowers annualized Sharpe ratios by 0.11, 0.19, and 0.22 (1.69%, 1.81%, and 2.15% a year at the factors' volatility), not significantly at the 5% level; gains above 0.07, 0.04, and 0.01 are rejected. Third, capping the combinations' weights at 2 leaves the horse-race verdict unchanged. Fourth, the signal is a weak variance forecast (slopes 0.48, 0.22, and 0.44; R2R^2 of 0.08, 0.01, and 0.07); weighting by its real-time forecast removes most of the loss (Sharpe differences of −0.04, 0.01, and −0.01) without a detectable gain, and Sharpe ratios differ by prior-year return (0.51, 0.55, and 0.51, significant at the 5% level or better) but not detectably by variance state.

We thus run the converse of the decomposition in Kim, Tse, and Wald (2016): in sample the timing value in these factors is momentum, not monthly-measured volatility, though in real time no momentum-based rule beats buy-and-hold at the 10% level. Noise in the monthly variance explains much of VM's loss, and its forecastable part has no detectable timing value; none of this bears directly on the daily-data rule.

Sections II to VI give the hypotheses, data and methods, results, robustness checks, and conclusions.

II. Hypothesis Development

A mean-variance investor holding wtw_t units of factor ff in month t+1t+1 chooses wt=μt/(γσt2)w_t=\mu_t/(\gamma\sigma_t^2). If the conditional mean μt\mu_t is constant, wt∝1/σt2w_t\propto 1/\sigma_t^2 and volatility management is optimal; if the conditional Sharpe ratio μt/σt\mu_t/\sigma_t is constant, wt∝1/σtw_t\propto 1/\sigma_t; if μt\mu_t rises one-for-one with σt2\sigma_t^2, timing has no value. Inverse-variance scaling pays only if the conditional Sharpe ratio falls when variance rises (Moreira and Muir 2017; Lochstoer and Muir 2022), and only if the estimated variance forecasts next-month variance, since the argument treats σt2\sigma_t^2 as known; with an estimated variance the weight should invert its forecast, as inverting a noisy estimate over-reacts to the noise. Because negative shocks raise subsequent volatility, the inverse-variance weight is also low after losses, so part of a volatility-managed alpha may be momentum timing. We test:

  • H1 (benchmark). Over the full sample, the managed factor earns a positive alpha against its unmanaged factor.
  • H2 (real time). With the scaling constant estimated from past data only, the managed factor's out-of-sample Sharpe ratio exceeds the unmanaged factor's, with a 95% interval excluding zero.
  • H3 (horse race). The managed alpha stays positive at the 5% level with TSFM in the spanning regression (H3a), and adding the volatility signal to a real-time momentum rule raises the out-of-sample Sharpe ratio (H3b).

VM survives for a factor if H2 and H3a both hold, and the hypothesis is supported if it survives for at least two factors. A failure of H2 is a refutation only if its interval lies entirely below zero, and inconclusive otherwise.

III. Data and Methodology

A. Data

We use the monthly Fama/French three-factor file of the Kenneth R. French Data Library, Tuck School of Business, Dartmouth College, built from CRSP and Compustat as in Fama and French (1993) and retrieved on 24 October 2026 as F-F_Research_Data_Factors_CSV.zip (SHA-256 prefix b840dba55d319f48). It holds 1,201 monthly observations, July 1926 to July 2026, of Mkt-RF (the value-weighted market return in excess of the one-month Treasury bill rate), SMB (small minus big), HML (high minus low book-to-market), and RF, in percent per month; we use the three factors and drop no observation, as there are no missing or sentinel values or calendar gaps. The library's daily file, needed for the Moreira-Muir signal, is not part of our data. Because the signals need 12 months of history, the in-sample tests use 1,189 months (July 1927 to July 2026) and the real-time evaluation 949 months from July 1947, after a 240-month initial window.

B. Signals and Strategies

At the end of month tt we compute the trailing six-month realized variance, the compounded prior-year return, and the momentum position,

RVt=16∑j=05ft−j2,mt=∏j=011(1+ft−j100)−1,st={+1,mt>0,−1,mt≤0,(1)RV_t=\frac{1}{6}\sum_{j=0}^{5}f_{t-j}^{2},\qquad m_t=\prod_{j=0}^{11}\Big(1+\frac{f_{t-j}}{100}\Big)-1,\qquad s_t=\begin{cases}+1, & m_t>0,\\ -1, & m_t\le 0,\end{cases}\tag{1}

and the strategies earn, in month t+1t+1,

VMt+1=cRVt ft+1,TSFMt+1=st ft+1.(2)VM_{t+1}=\frac{c}{RV_t}\,f_{t+1},\qquad TSFM_{t+1}=s_t\,f_{t+1}.\tag{2}

In sample, cc equates the full-sample volatilities of VMVM and ff, as in Moreira and Muir (2017); it rescales the alpha but leaves t-statistics and Sharpe ratios unchanged. In real time we use

ct=σ^(fτ:τ≤t)σ^(fτ/RVτ−1:τ≤t),(3)c_t=\frac{\hat\sigma\left(f_\tau:\tau\le t\right)}{\hat\sigma\left(f_\tau/RV_{\tau-1}:\tau\le t\right)},\tag{3}

estimated on an expanding window from July 1927, so the weight ct/RVtc_t/RV_t uses no later information. A capped variant limits the weight to 2.

C. Spanning Regressions

We estimate

VMt+1=α+βft+1+δ TSFMt+1+εt+1(4)VM_{t+1}=\alpha+\beta f_{t+1}+\delta\,TSFM_{t+1}+\varepsilon_{t+1}\tag{4}

with and without the TSFM term, and the reverse regressions of TSFMTSFM on ff and VMVM. A positive α\alpha means that the left-hand portfolio expands the frontier of the right-hand ones (Huberman and Kandel 1987). Alphas are in percent per year, with Newey and West (1987) standard errors (12 lags). Joint tests that the three factors' alphas are zero are Wald tests on the stacked regressions with a Newey-West (12 lags) covariance, χ2\chi^2 with three degrees of freedom.

D. Real-Time Evaluation and Diagnostics

At the end of each month from June 1947 we estimate the mean μ^t\hat\mu_t and covariance Σ^t\hat\Sigma_t of the realized returns fτ (1, 1/RVτ−1, sτ−1)′f_\tau\,(1,\,1/RV_{\tau-1},\,s_{\tau-1})', τ≤t\tau\le t, and form

θt=Σ^t−1μ^t,Wt=θt′ (1, 1/RVt, st)′(5)\theta_t=\hat\Sigma_t^{-1}\hat\mu_t,\qquad W_t=\theta_t'\,(1,\,1/RV_t,\,s_t)'\tag{5}

for {f, VM}, {f, TSFM}, and {f, VM, TSFM}, dropping unused elements and rescaling θt\theta_t to the estimated variance of ff; the combination earns Wtft+1W_t f_{t+1}. Because sample mean-variance weights are prone to estimation error (DeMiguel, Garlappi, and Uppal 2009), we also cap ∣Wt∣|W_t| at 2. Sharpe-ratio differences Δ^=SR^a−SR^b\hat\Delta=\widehat{SR}_a-\widehat{SR}_b are tested with the delta method that descends from Jobson and Korkie (1981) and Lo (2002), in the form robust to heavy tails and serial dependence,

Var^(Δ^)=T−1 ∇Δ^′ Ψ^ ∇Δ^,(6)\widehat{\mathrm{Var}}(\hat\Delta)=T^{-1}\,\nabla\hat\Delta'\,\hat\Psi\,\nabla\hat\Delta,\tag{6}

where Ψ^\hat\Psi is the Newey-West (Bartlett, 12 lags) long-run covariance of (ra,rb,ra2,rb2)(r_a,r_b,r_a^2,r_b^2) (Ledoit and Wolf 2008). The headline difference also gets a circular block-bootstrap percentile interval (block length 12, 2,000 draws, seed 20261024), and the same delta method gives the mean-return gap, in percent per year, of a strategy scaled to its factor's 1947 to 2026 volatility. For each weight WtW_t we report leverage, turnover ∣Wt−Wt−1∣|W_t-W_{t-1}|, and the break-even one-way cost κ∗\kappa^* solving

SR(Wtft+1−κ∗∣Wt−Wt−1∣)=SR(f).(7)SR\left(W_t f_{t+1}-\kappa^*|W_t-W_{t-1}|\right)=SR(f).\tag{7}

To test the signal, over July 1927 to July 2026 we estimate

ft+12=a+b RVt+et+1,ft+1=a′+b′ RVt−RV‾sRV+et+1′.(8)f_{t+1}^2=a+b\,RV_t+e_{t+1},\qquad f_{t+1}=a'+b'\,\frac{RV_t-\overline{RV}}{s_{RV}}+e'_{t+1}.\tag{8}

An unbiased variance forecast has b=1b=1; b′b' is the return effect, in percent per month, of a one-standard-deviation rise in RVtRV_t (full-sample standardization only rescales it). Months are high variance when RVtRV_t exceeds the median of RVRV from December 1926 through month tt; state means and Sharpe ratios are compared by regression and the delta method. To test whether noise in RVtRV_t drives the results, we also weight each factor in real time by

wtF=ctFa^t+b^t RVt,(9)w^F_t=\frac{c^F_t}{\hat a_t+\hat b_t\,RV_t},\tag{9}

where a^t\hat a_t and b^t\hat b_t are OLS estimates of the variance equation in (8) from July 1927 through month tt, and ctFc^F_t is equation (3) with RVτ−1RV_{\tau-1} replaced by a^t+b^t RVτ−1\hat a_t+\hat b_t\,RV_{\tau-1}, so this weight, too, is real time.

IV. Results

A. Summary Statistics

Table I shows that the market earns 8.26% a year with a volatility of 18.44% and a Sharpe ratio of 0.45; size and value earn 2.07% and 4.19%. In sample VM lowers the Sharpe ratio by 0.07, 0.19, and 0.17 (significant at the 5% level for size and 10% for value; a constant cannot change a Sharpe ratio, so there is no look-ahead). The volatility weight and the momentum position are positively correlated for the market (0.26; mean managed weight 1.27 when TSFM is long, 0.68 when short), negatively for size (−0.17), and not for value (0.01).

Table I. Summary Statistics, July 1927 to July 2026

In-sample statistics, NN = 1,189 months. VM is ff scaled by the "VM weight" c/RVtc/RV_t with the full-sample cc; TSFM is stft+1s_t f_{t+1}.

Mkt-RFSMBHML
Mean (% per year)8.262.074.19
Standard deviation (% per year)18.4410.9512.33
Sharpe ratio, unmanaged factor0.45*** [0.22, 0.68]0.19* [−0.01, 0.39]0.34*** [0.13, 0.55]
Sharpe ratio, VM0.38*** [0.15, 0.62]0.00 [−0.22, 0.22]0.17 [−0.04, 0.37]
Sharpe ratio, TSFM0.32*** [0.12, 0.51]0.30*** [0.10, 0.50]0.30*** [0.11, 0.50]
SR(VM) − SR(ff)−0.07 [−0.26, 0.13]−0.19** [−0.36, −0.01]−0.17* [−0.37, 0.02]
SR(TSFM) − SR(ff)−0.13 [−0.41, 0.15]0.11 [−0.14, 0.36]−0.04 [−0.30, 0.23]
Share of months with st=+1s_t=+10.710.530.61
Correlation of 1/RVt1/RV_t with sts_t0.26−0.170.01
Mean VM weight, st=+1s_t=+11.270.811.05
Mean VM weight, st=−1s_t=-10.681.251.01

Notes: Sharpe ratios and differences are annualized, with 95% delta-method intervals (Newey-West, 12 lags). *, **, and *** denote significance at the 10%, 5%, and 1% levels.

B. Spanning Regressions and the In-Sample Horse Race

In Panel A of Table II, H1 fails for every factor: the market's alpha is 1.64% a year (t = 0.90; interval −1.92% to 5.20%), the size and value alphas are −1.17% and 0.14%, and the three are jointly insignificant (p = 0.37); Section V re-estimates H1 under every alternative specification. With TSFM added the market alpha remains indistinguishable from zero (−0.36%, t = −0.23), so there is no alpha to decompose; the content is the momentum loading of 0.31 (t = 5.64), which raises the adjusted R2R^2 from 0.43 to 0.52. H3a fails for all three factors (joint p = 0.85). The informative direction is Panel B: TSFM alphas of 6.42% (t = 2.97), 2.75% (t = 2.61), and 3.23% (t = 2.86) a year barely move with VM as a control (5.52%, 2.54%, and 3.24%) and remain jointly significant at the 1% level.

Table II. Spanning Regressions and the In-Sample Horse Race, July 1927 to July 2026

OLS estimates of equation (4) and its reverse; NN = 1,189 in every regression.

Panel A. Dependent variable: VMt+1VM_{t+1}

FactorRegressorsα\alpha, % per year (tt)95% CI for α\alphaβ\beta on ff (tt)δ\delta on TSFM (tt)Adj. R2R^2
Mkt-RFff1.64 (0.90)[−1.92, 5.20]0.65*** (6.39)—0.43
Mkt-RFff, TSFM−0.36 (−0.23)[−3.45, 2.72]0.67*** (9.49)0.31*** (5.64)0.52
SMBff−1.17 (−1.25)[−3.00, 0.67]0.58*** (7.37)—0.33
SMBff, TSFM−0.81 (−0.86)[−2.65, 1.04]0.61*** (9.06)−0.13** (−2.07)0.35
HMLff0.14 (0.13)[−1.96, 2.24]0.45*** (4.58)—0.21
HMLff, TSFM0.19 (0.18)[−1.91, 2.29]0.46*** (4.66)−0.02 (−0.30)0.21

Panel B. Dependent variable: TSFMt+1TSFM_{t+1}

FactorRegressorsα\alpha, % per year (tt)95% CI for α\alphaβ\beta on ff (tt)δ\delta on VM (tt)Adj. R2R^2
Mkt-RFff6.42*** (2.97)[2.18, 10.67]−0.07 (−0.44)—0.00
Mkt-RFff, VM5.52*** (3.13)[2.07, 8.98]−0.42*** (−2.59)0.55*** (5.34)0.17
SMBff2.75*** (2.61)[0.69, 4.81]0.24 (1.62)—0.06
SMBff, VM2.54** (2.38)[0.45, 4.62]0.35* (1.91)−0.18** (−2.00)0.08
HMLff3.23*** (2.86)[1.02, 5.45]0.12 (0.94)—0.01
HMLff, VM3.24*** (2.87)[1.03, 5.45]0.13 (0.86)−0.02 (−0.30)0.01

Notes: tt-statistics (parentheses) and normal-based intervals use Newey-West standard errors with 12 lags. HAC Wald statistics (3 degrees of freedom) for the three alphas being jointly zero: 3.13 (pp = 0.37) and 0.78 (pp = 0.85) in Panel A, 24.93 and 23.81 (both pp < 0.001) in Panel B. *, **, and *** denote significance at the 10%, 5%, and 1% levels.

C. Real-Time Evaluation

In Table III the real-time managed factors trail their factors from July 1947 by 0.11 in Sharpe ratio for the market (interval −0.29 to 0.07), 0.19 for size (−0.41 to 0.04), and 0.22 for value (−0.45 to 0.01; significant at the 10% level only); the bootstrap intervals are nearly identical. At the factor's volatility (Panel D), VM earns 1.69%, 1.81%, and 2.15% a year less, with intervals that include zero. H2 fails for every factor, and gains above 0.07, 0.04, and 0.01 in annualized Sharpe ratio are rejected, a generous bound as returns are gross. With H3a also failing, VM survives for none of the three factors, so under our pre-committed rule the hypothesis is not supported; only the narrower proposition that real-time VM is worse than its factor is left inconclusive, because no baseline interval lies entirely below zero.

A positive constant cannot change a Sharpe ratio, so the full-sample-cc row equals the Sharpe ratio of ft+1/RVtf_{t+1}/RV_t by construction, and its comparison with the real-time row (−0.13 against −0.11 for the market) measures only the time variation in ctc_t; of Liu, Tang, and Zhou's (2019) critique about leverage, drawdowns, and alphas, we address only leverage (Table V). Capping the VM weight at 2 narrows the shortfall for the market and value but makes it significant at the 5% level for size (−0.17).

In the real-time horse race, adding VM to a momentum rule lowers the Sharpe ratio by 0.03, 0.03, and 0.06, so H3b fails, and adding momentum to VM raises it by 0.05, 0.10, and 0.08. With combination weights capped at 2 (Panel C; binding in up to a tenth of months), the H3b differences are −0.03, 0.01, and −0.01. No horse-race difference, and no timing rule's gain over buy-and-hold, is significant at the 10% level (the largest gain, 0.14 for size, is TSFM's), and neither of the two Panel B differences significant at the 5% level individually survives a Holm adjustment (smallest adjusted p = 0.92). In-sample momentum alphas thus shrink in real time, in line with Huang et al. (2020), Cederburg et al. (2020), and Goyal and Welch (2008).

Table III. Real-Time Out-of-Sample Performance, July 1947 to July 2026

Annualized Sharpe ratios from data through month tt only, except the full-sample-cc reference row; NN = 949. Panel C caps combination weights at ∣Wt∣≤2|W_t|\le 2 (never binding for {f, TSFM}); Panel D scales strategies to the factor's volatility.

Panel A. Sharpe ratios

Mkt-RFSMBHML
Unmanaged factor ff0.56*** [0.30, 0.81]0.12 [−0.12, 0.35]0.38*** [0.11, 0.64]
VM, real-time ctc_t0.44*** [0.19, 0.69]−0.07 [−0.32, 0.18]0.16 [−0.06, 0.38]
TSFM0.34*** [0.12, 0.55]0.26** [0.04, 0.48]0.31** [0.07, 0.54]
{f, VM}0.51*** [0.26, 0.76]0.11 [−0.13, 0.34]0.31** [0.06, 0.56]
{f, TSFM}0.59*** [0.37, 0.82]0.24** [0.02, 0.47]0.44*** [0.21, 0.67]
{f, VM, TSFM}0.56*** [0.34, 0.78]0.21* [−0.02, 0.44]0.38*** [0.14, 0.62]

Panel B. Differences in Sharpe ratios

Mkt-RFSMBHML
VM (real time) − ff (H2)−0.11 [−0.29, 0.07]−0.19 [−0.41, 0.04]−0.22* [−0.45, 0.01]
block-bootstrap percentile 95% CI[−0.29, 0.07][−0.41, 0.02][−0.44, 0.02]
VM (full-sample cc) − ff−0.13 [−0.31, 0.05]−0.17 [−0.38, 0.04]−0.23** [−0.46, −0.00]
VM (capped) − ff−0.09 [−0.23, 0.06]−0.17** [−0.34, −0.01]−0.10 [−0.30, 0.09]
TSFM − ff−0.22 [−0.52, 0.08]0.14 [−0.15, 0.43]−0.07 [−0.41, 0.26]
{f, VM} − ff−0.04 [−0.15, 0.06]−0.01 [−0.20, 0.18]−0.07 [−0.17, 0.03]
{f, TSFM} − ff0.03 [−0.12, 0.19]0.13 [−0.10, 0.36]0.07 [−0.10, 0.23]
{f, VM, TSFM} − ff0.00 [−0.16, 0.16]0.09 [−0.16, 0.35]0.00 [−0.17, 0.18]
{f, VM, TSFM} − {f, TSFM} (H3b)−0.03 [−0.08, 0.01]−0.03 [−0.15, 0.09]−0.06 [−0.15, 0.03]
{f, VM, TSFM} − {f, VM}0.05 [−0.09, 0.18]0.10 [−0.09, 0.29]0.08 [−0.05, 0.20]

Panel C. Differences in Sharpe ratios, combination weights capped at 2

Mkt-RFSMBHML
{f, VM} − ff−0.03 [−0.10, 0.04]0.04 [−0.08, 0.16]−0.03 [−0.08, 0.02]
{f, VM, TSFM} − ff0.01 [−0.15, 0.16]0.13 [−0.11, 0.37]0.05 [−0.11, 0.21]
{f, VM, TSFM} − {f, TSFM} (H3b)−0.03 [−0.06, 0.01]0.01 [−0.05, 0.06]−0.01 [−0.07, 0.04]
{f, VM, TSFM} − {f, VM}0.04 [−0.10, 0.18]0.09 [−0.12, 0.30]0.08 [−0.06, 0.23]
Share of months at the cap, {f, VM} / {f, VM, TSFM}0.10 / 0.030.02 / 0.020.08 / 0.07

Panel D. Mean-return gap at the factor's volatility (% per year)

Mkt-RFSMBHML
VM (real time) − ff−1.69 [−4.38, 0.99]−1.81 [−3.99, 0.38]−2.15* [−4.45, 0.15]
VM (capped) − ff−1.31 [−3.46, 0.84]−1.68** [−3.27, −0.09]−1.01 [−2.92, 0.89]
TSFM − ff−3.27 [−7.62, 1.07]1.39 [−1.37, 4.15]−0.71 [−4.00, 2.58]
{f, VM} − ff−0.66 [−2.23, 0.90]−0.10 [−1.93, 1.73]−0.70 [−1.68, 0.28]
{f, TSFM} − ff0.50 [−1.84, 2.84]1.23 [−0.96, 3.42]0.65 [−0.97, 2.27]
{f, VM, TSFM} − ff0.01 [−2.36, 2.38]0.90 [−1.56, 3.37]0.04 [−1.67, 1.75]

Notes: 95% delta-method intervals (Ledoit and Wolf 2008; Newey-West, 12 lags); the bootstrap row is a circular block-bootstrap percentile interval (block 12, 2,000 draws, seed 20261024). The HML upper limit in the full-sample-cc row is −0.0015. *, **, *** denote the 10%, 5%, 1% levels. Returns are gross.

D. Why Monthly Volatility Management Fails

Panel C of Table IV tests the signal. The six-month RVtRV_t forecasts next-month squared returns with slopes of 0.48 (t = 4.67), 0.22 (t = 2.09), and 0.44 (t = 3.95), below one at the 1% level (t of −4.99, −7.38, and −4.95), with R2R^2 of 0.08, 0.01, and 0.07. Such slopes are consistent with a noisy signal, and the twelve-month window, with slopes closest to one (0.75, 0.34, and 0.59), has the smallest real-time shortfall of the three windows for the market (Table VII). A one-standard-deviation rise in the six-month RVtRV_t moves next-month returns by 0.11%, −0.08%, and −0.13% a month, and no window's slope is significant at the 10% level.

Panel D tests the noisy-signal explanation directly with the weight of equation (9), which inverts the real-time forecast rather than RVtRV_t. With slope estimates of 0.03 to 0.49 it moves little, peaking at 1.86, 1.29, and 1.86 against 12.10, 18.46, and 37.95 for ct/RVtc_t/RV_t (Table V). It beats real-time VM by 0.07, 0.19, and 0.21 in Sharpe ratio (significant at the 10% level for size and 5% for value, not for the market) and differs from the factor by only −0.04, 0.01, and −0.01 (−0.63%, 0.06%, and −0.14% a year at matched volatility), none significant at the 10% level; gains above 0.02, 0.05, and 0.07 are rejected. Inverting the noisy estimate rather than its forecast thus accounts for most of VM's measured loss, most clearly for value, but once the weight reflects how little variance is forecastable, any remaining timing value is too small to detect.

Panel A is consistent with this. Across variance states conditional Sharpe ratios differ by −0.01, 0.20, and 0.05, with intervals about ±0.4 wide: the data cannot tell whether reward per unit of risk changes with variance. Equal Sharpe ratios favor 1/σ1/\sigma timing only if σt\sigma_t is known, and with forecast slopes well below one even inverse-volatility weights swing far more than forecastable risk; their real-time differences are −0.05, −0.10, and −0.07, none significant at the 5% level (Table VII). The prior-year return separates reward sharply: Sharpe ratios after a positive year exceed those after a negative one by 0.51, 0.55, and 0.51 (significant at the 5%, 1%, and 5% levels), and mean returns by about 50 basis points a month, as in Ehsani and Linnainmaa (2022). The detectable timing information in these data is in past returns, not past variance.

Table IV. States, the Variance Signal, and a Forecast-Based Weight

Panel A splits months by whether RVtRV_t exceeds its expanding median (from December 1926), Panel B by the sign of mtm_t, and Panel C estimates equation (8), all over July 1927 to July 2026 (NN = 1,189); Panel D evaluates the real-time weight of equation (9) over July 1947 to July 2026 (NN = 949). Returns are for month t+1t+1.

Panel A. Variance state

Mkt-RFSMBHML
Mean, high variance (% per month)0.880.300.51
Mean, low variance (% per month)0.560.050.22
Difference in means (tt)0.32 (0.91)0.26 (1.40)0.29 (1.24)
95% CI for the difference in means[−0.37, 1.01][−0.10, 0.62][−0.17, 0.75]
Sharpe ratio, high variance / low variance0.45 / 0.470.27 / 0.070.38 / 0.33
Difference in Sharpe ratios [95% CI]−0.01 [−0.46, 0.43]0.20 [−0.18, 0.58]0.05 [−0.35, 0.44]

Panel B. Momentum state

Mkt-RFSMBHML
Mean, after mt>0m_t>0 (% per month)0.830.420.54
Mean, after mt≤0m_t\le 0 (% per month)0.34−0.110.05
Difference in means (tt)0.50 (1.26)0.52*** (2.82)0.49** (2.46)
95% CI for the difference in means[−0.28, 1.27][0.16, 0.89][0.10, 0.88]
Sharpe ratio, after mt>0m_t>0 / after mt≤0m_t\le 00.67 / 0.160.43 / −0.130.55 / 0.05
Difference in Sharpe ratios [95% CI]0.51** [0.07, 0.94]0.55*** [0.16, 0.95]0.51** [0.11, 0.90]

Panel C. Forecasting next-month variance and return with RVtRV_t

FactorWindowSlope of ft+12f_{t+1}^2 on RVtRV_t (tt)tt for slope = 1R2R^2One-SD effect of RVtRV_t on ft+1f_{t+1}, % per month (tt)R2R^2
Mkt-RF3 months0.45*** (4.99)−5.980.100.32 (1.43)0.00
Mkt-RF6 months0.48*** (4.67)−4.990.080.11 (0.57)0.00
Mkt-RF12 months0.75*** (4.32)−1.400.150.32 (1.13)0.00
SMB3 months0.16* (1.68)−8.770.01−0.04 (−0.30)0.00
SMB6 months0.22** (2.09)−7.380.01−0.08 (−0.84)0.00
SMB12 months0.34*** (3.21)−6.140.020.13 (1.17)0.00
HML3 months0.39*** (4.46)−7.000.08−0.07 (−0.41)0.00
HML6 months0.44*** (3.95)−4.950.07−0.13 (−0.92)0.00
HML12 months0.59*** (3.98)−2.810.080.24 (1.28)0.00

Panel D. Factor weighted by the real-time variance forecast, equation (9)

Mkt-RFSMBHML
Sharpe ratio0.51*** [0.26, 0.77]0.12 [−0.13, 0.37]0.36*** [0.10, 0.63]
Difference from ff−0.04 [−0.11, 0.02]0.01 [−0.03, 0.05]−0.01 [−0.10, 0.07]
Difference from VM, real time0.07 [−0.07, 0.21]0.19* [−0.03, 0.42]0.21** [0.01, 0.40]
Mean-return gap to ff at the factor's volatility (% per year)−0.63 [−1.59, 0.33]0.06 [−0.33, 0.45]−0.14 [−0.99, 0.70]
Mean / maximum weight1.35 / 1.861.07 / 1.291.43 / 1.86
Turnover per month0.080.020.07
Range of b^t\hat b_t0.40 to 0.490.03 to 0.250.38 to 0.45

Notes: Mean differences are slopes on a state dummy (Newey-West, 12 lags); Sharpe-ratio differences and their intervals use equation (6). In Panel C the stars test each slope against zero, and the column "tt for slope = 1" tests the variance slope against one. *, **, and *** denote the 10%, 5%, and 1% levels. High-variance months: 481, 590, 521; positive-momentum months: 840, 629, 727 (Mkt-RF, SMB, HML).

E. Leverage, Turnover, and Costs

In Table V the uncapped real-time managed factors average leverage of 1.26 to 1.47, exceed a weight of 2 in 16.7% to 21.6% of months, peak at 12.10, 18.46, and 37.95, and turn over 0.36 to 0.46 a month, with no gross gain to break even on; {f, VM} reaches absolute weights of 7.14, 16.01, and 15.44. TSFM and {f, TSFM} trade less, and their gross gains, not significant at the 10% level, survive one-way costs of 53.92 basis points (TSFM, size) and 46, 53, and 47 basis points ({f, TSFM}) per unit of turnover (see Novy-Marx and Velikov 2016).

Table V. Leverage, Turnover, and Break-Even Costs, July 1947 to July 2026

Statistics of the weight WtW_t each real-time strategy places on the factor; NN = 949 months.

FactorStrategyMean weightMaximum absolute weightShare of months with absolute weight above 2Turnover per monthBreak-even cost (bp)
Mkt-RFVM, real time1.4712.100.220.37n.a.
Mkt-RFVM, capped at 21.182.000.000.16n.a.
Mkt-RFTSFM0.461.000.000.15n.a.
Mkt-RF{f, VM}1.347.140.100.17n.a.
Mkt-RF{f, TSFM}1.171.720.000.1146.45
Mkt-RF{f, VM, TSFM}1.184.090.030.170.59
SMBVM, real time1.2618.460.180.36n.a.
SMBVM, capped at 20.932.000.000.15n.a.
SMBTSFM−0.031.000.000.2153.92
SMB{f, VM}0.5016.010.020.21n.a.
SMB{f, TSFM}0.341.230.000.1952.56
SMB{f, VM, TSFM}0.1411.510.020.2728.70
HMLVM, real time1.3237.950.170.46n.a.
HMLVM, capped at 20.952.000.000.17n.a.
HMLTSFM0.231.000.000.19n.a.
HML{f, VM}1.3015.440.080.15n.a.
HML{f, TSFM}0.861.360.000.1147.38
HML{f, VM, TSFM}1.1314.280.070.231.73

Notes: Turnover is the mean of ∣Wt−Wt−1∣|W_t-W_{t-1}|. The break-even cost solves equation (7), in basis points per unit of turnover; "n.a." means the gross Sharpe ratio does not exceed the factor's.

F. Weight Paths and Cumulative Returns

Table VI gives the paths in place of figures. At matched volatility the managed market factor built its lead in 1947 to 1966, at mean weights of 1.82 and 2.52, lost it by 1986, and ended at 522.94% against 656.91% for the factor; for size and value VM ended at −53.73% and 121.97% (factors 89.14% and 291.85%). The early high weights are consistent with a constant estimated over the Depression's extreme volatility (Schwert 1989).

Table VI. Real-Time Weights and Cumulative Returns by Decade, July 1947 to July 2026

Decade mean and maximum of the real-time VM weight ct/RVtc_t/RV_t, share of months TSFM is long, and cumulative sums of monthly percentage returns from July 1947, with VM and TSFM rescaled after the fact to the factor's 1947 to 2026 volatility.

Panel A. Mkt-RF

DecadeMean VM weightMaximum VM weightShare of months TSFM longCumulative ff (%)Cumulative VM (%)Cumulative TSFM (%)
1947–19561.828.420.89150.22178.66103.05
1957–19662.529.520.64218.37264.67115.28
1967–19761.135.310.59237.23279.52168.83
1977–19861.063.580.66299.88251.76170.27
1987–19961.595.570.78395.71334.06208.09
1997–20061.195.020.73455.84388.39287.90
2007–20161.244.120.77532.71441.47358.83
2017–20261.2412.100.83656.91522.94398.27

Panel B. SMB

DecadeMean VM weightMaximum VM weightShare of months TSFM longCumulative ff (%)Cumulative VM (%)Cumulative TSFM (%)
1947–19562.769.830.21−24.93−61.8211.42
1957–19662.6318.460.57−6.22−37.8637.00
1967–19760.493.150.5515.88−46.28119.41
1977–19861.085.180.7283.34−35.50209.35
1987–19960.823.360.3868.57−51.58219.54
1997–20060.431.670.6098.82−37.92225.43
2007–20161.205.100.52113.39−44.16198.72
2017–20260.693.430.3389.14−53.73198.73

Panel C. HML

DecadeMean VM weightMaximum VM weightShare of months TSFM longCumulative ff (%)Cumulative VM (%)Cumulative TSFM (%)
1947–19562.9718.670.6443.1557.6221.37
1957–19662.1237.950.7181.5694.1357.38
1967–19760.855.450.68153.76116.14103.49
1977–19860.893.220.68212.79113.10137.11
1987–19961.3010.660.59239.7394.83133.72
1997–20060.906.400.75314.03139.67216.23
2007–20161.135.780.38304.28130.45212.84
2017–20260.476.380.50291.85121.97235.81

Notes: Annualized 1947 to 2026 volatility of the unscaled VM and TSFM against the factor: Mkt-RF 22.42% and 15.02% against 14.90%; SMB 12.53% and 9.64% against 9.66%; HML 15.81% and 9.79% against 9.77%.

V. Robustness

Table VII varies one design choice at a time. In Panel A, four of the 24 cells in the eight specification rows are significant at the 5% level alone, none after a Holm adjustment (smallest adjusted p = 0.06), so no cell should be read in isolation. A three-month window widens the shortfall for the market (−0.27) and value (−0.36); a twelve-month window (−0.05, −0.15, −0.14) and inverse-volatility scaling (−0.05, −0.10, −0.07; Harvey et al. 2018) narrow it, and the forecast weight nearly closes it (−0.04, 0.01, −0.01). Of the 24 specification differences, 23 are negative and none is positive at the 10% level; the largest shortfall is −0.40 (value, 1947 to 1986). In 2016:01 to 2026:07, which largely postdates Moreira and Muir (2017), the estimates (−0.21, 0.13, and −0.05) have intervals as wide as −0.80 to 0.38, too wide to detect the post-publication decay McLean and Pontiff (2016) document.

Panel B re-estimates H1 in every specification. One of its 24 alphas is significant at the 5% level, the market's with a twelve-month window: 3.35% a year (t = 2.12; interval 0.26% to 6.44%). But the three twelve-month alphas are not jointly significant (p = 0.10), no other alpha is significant at the 10% level, the joint tests' p-values run from 0.10 to 0.93, and across the 24 overlapping cells this one's Holm-adjusted p is 0.81. The twelve-month window is also where the signal best forecasts the market's variance (slope 0.75); the alpha halves to 1.64% (interval −0.93% to 4.21%) once TSFM enters (Panel C), and its real-time difference is −0.05, so VM does not survive there either. No horse-race alpha in Panel C is significant at the 10% level, singly or jointly (joint p from 0.33 to 0.94).

Table VII. Robustness of the Headline Statistics

Panel A: real-time SR(VM) − SR(ff), July 1947 to July 2026 or the stated subperiod; its last row replaces ct/RVtc_t/RV_t with the forecast weight of equation (9). Panels B and C: in-sample α\alpha (% per year) of VMVM on ff (H1) and on ff and TSFM (H3a), July 1927 to July 2026 or the subperiod; subperiod alphas use the full-sample cc, which only rescales them. Joint pp: HAC Wald test that the three alphas are zero.

Panel A. Real-time Sharpe-ratio difference

SpecificationNNMkt-RFSMBHML
Baseline (six-month RVRV, 1/RV1/RV)949−0.11 [−0.29, 0.07]−0.19 [−0.41, 0.04]−0.22* [−0.45, 0.01]
Three-month RVRV949−0.27*** [−0.48, −0.07]−0.01 [−0.27, 0.25]−0.36** [−0.66, −0.07]
Twelve-month RVRV949−0.05 [−0.21, 0.10]−0.15 [−0.32, 0.03]−0.14 [−0.34, 0.07]
1/RV1/\sqrt{RV} (volatility targeting)949−0.05 [−0.15, 0.06]−0.10* [−0.21, 0.01]−0.07 [−0.21, 0.08]
Weight capped at 2949−0.09 [−0.23, 0.06]−0.17** [−0.34, −0.01]−0.10 [−0.30, 0.09]
1947:07–1986:12474−0.14 [−0.36, 0.09]−0.32* [−0.68, 0.05]−0.40*** [−0.66, −0.14]
1987:01–2026:07475−0.08 [−0.35, 0.20]−0.08 [−0.30, 0.13]−0.15 [−0.53, 0.22]
2016:01–2026:07127−0.21 [−0.80, 0.38]0.13 [−0.21, 0.47]−0.05 [−0.67, 0.58]
Forecast weight, equation (9)949−0.04 [−0.11, 0.02]0.01 [−0.03, 0.05]−0.01 [−0.10, 0.07]

Panel B. Benchmark alpha (H1)

SpecificationNNMkt-RFSMBHMLJoint pp
Baseline (six-month RVRV, 1/RV1/RV)1,1891.64 [−1.92, 5.20]−1.17 [−3.00, 0.67]0.14 [−1.96, 2.24]0.37
Three-month RVRV1,1891.31 [−2.34, 4.96]0.96 [−0.98, 2.91]−0.75 [−3.23, 1.73]0.61
Twelve-month RVRV1,1893.35** [0.26, 6.44]−0.79 [−2.38, 0.80]0.67 [−1.38, 2.71]0.10
1/RV1/\sqrt{RV} (volatility targeting)1,1891.25 [−1.06, 3.55]−0.67 [−1.71, 0.36]0.63 [−0.94, 2.20]0.25
Weight capped at 21,1890.96 [−1.77, 3.69]−0.83 [−1.94, 0.27]0.49 [−0.86, 1.84]0.27
1947:07–1986:12474−1.42 [−5.31, 2.46]−2.38 [−6.25, 1.49]−2.06 [−4.91, 0.79]0.27
1987:01–2026:074751.64 [−3.69, 6.96]−0.75 [−2.61, 1.11]−0.74 [−4.48, 3.00]0.65
2016:01–2026:071271.27 [−12.67, 15.20]0.75 [−1.47, 2.97]−0.26 [−4.97, 4.45]0.93

Panel C. Horse-race alpha (H3a)

SpecificationNNMkt-RFSMBHMLJoint pp
Baseline (six-month RVRV, 1/RV1/RV)1,189−0.36 [−3.45, 2.72]−0.81 [−2.65, 1.04]0.19 [−1.91, 2.29]0.85
Three-month RVRV1,1890.12 [−3.38, 3.63]0.94 [−0.86, 2.74]−0.77 [−3.19, 1.65]0.68
Twelve-month RVRV1,1891.64 [−0.93, 4.21]−0.29 [−1.85, 1.27]1.02 [−1.07, 3.10]0.48
1/RV1/\sqrt{RV} (volatility targeting)1,189−0.27 [−2.10, 1.56]−0.39 [−1.45, 0.66]0.83 [−0.77, 2.44]0.63
Weight capped at 21,189−0.74 [−3.02, 1.54]−0.58 [−1.66, 0.50]0.59 [−0.76, 1.93]0.54
1947:07–1986:12474−2.27 [−5.91, 1.37]−1.63 [−5.73, 2.47]−1.88 [−4.83, 1.06]0.33
1987:01–2026:074750.42 [−4.21, 5.06]−0.80 [−2.62, 1.02]−0.65 [−4.37, 3.06]0.79
2016:01–2026:071272.04 [−10.62, 14.71]0.73 [−1.57, 3.04]−0.02 [−4.40, 4.36]0.94

Notes: 95% intervals: delta method (Panel A), Newey-West with 12 lags (Panels B and C). *, **, and *** denote the 10%, 5%, and 1% levels, unadjusted for multiple comparisons.

The conclusions apply to monthly volatility management, not the daily-data rule, and to three factors; momentum, profitability, and investment factors are not in our data. Returns ignore financing and margin costs, pre-1963 coverage is thinner, and the histories are the current vintage. The forecast weight uses one linear forecast from one window; we did not test richer variance forecasts, the one-month squared return, the reverse horse race beyond the baseline, other momentum look-backs, or net-of-cost alphas.

VI. Conclusion

Volatility management measured from monthly returns does not improve the Fama-French market, size, and value factors over 1926 to 2026. Its alphas are indistinguishable from zero with or without a momentum control; the one benchmark alpha of 24 significant at the 5% level, the market's with a twelve-month window, is not jointly significant with the size and value alphas (p = 0.10) and does not survive a Holm adjustment. Its real-time Sharpe ratios trail the factors' with intervals that exclude all but small gains. Inverting the noisy six-month variance rather than its real-time forecast accounts for most of that loss: the forecast weight beats inverse-variance weighting at the 5% level for value and the 10% level for size, though not significantly for the market, and adds no detectable gain over the factors. In sample the timing value is momentum, but in real time no momentum-based rule beats buy-and-hold at the 10% level. The decisive next step, which our data cannot support, is the same horse race with the daily-return realized variance of Moreira and Muir (2017), valued for risk forecasting and timing (Fleming, Kirby, and Ostdiek 2003; Bollerslev et al. 2018), and with the momentum factor, where volatility scaling is best documented (Barroso and Santa-Clara 2015; Daniel and Moskowitz 2016).

References

Arnott, Robert D., Vitali Kalesnik, and Juhani T. Linnainmaa, 2023, Factor momentum, Review of Financial Studies 36, 3034–3070. https://doi.org/10.1093/rfs/hhad006

Barroso, Pedro, and Andrew Detzel, 2021, Do limits to arbitrage explain the benefits of volatility-managed portfolios?, Journal of Financial Economics 140, 744–767. https://doi.org/10.1016/j.jfineco.2021.02.009

Barroso, Pedro, and Pedro Santa-Clara, 2015, Momentum has its moments, Journal of Financial Economics 116, 111–120. https://doi.org/10.1016/j.jfineco.2014.11.010

Bollerslev, Tim, Benjamin Hood, John Huss, and Lasse Heje Pedersen, 2018, Risk everywhere: Modeling and managing volatility, Review of Financial Studies 31, 2729–2773. https://doi.org/10.1093/rfs/hhy041

Campbell, John Y., and Ludger Hentschel, 1992, No news is good news: An asymmetric model of changing volatility in stock returns, Journal of Financial Economics 31, 281–318. https://doi.org/10.1016/0304-405X(92)90037-X

Cederburg, Scott, Michael S. O'Doherty, Feifei Wang, and Xuemin (Sterling) Yan, 2020, On the performance of volatility-managed portfolios, Journal of Financial Economics 138, 95–117. https://doi.org/10.1016/j.jfineco.2020.04.015

Daniel, Kent, and Tobias J. Moskowitz, 2016, Momentum crashes, Journal of Financial Economics 122, 221–247. https://doi.org/10.1016/j.jfineco.2015.12.002

DeMiguel, Victor, Lorenzo Garlappi, and Raman Uppal, 2009, Optimal versus naive diversification: How inefficient is the 1/N portfolio strategy?, Review of Financial Studies 22, 1915–1953. https://doi.org/10.1093/rfs/hhm075

DeMiguel, Victor, Alberto Martín-Utrera, and Raman Uppal, 2024, A multifactor perspective on volatility-managed portfolios, Journal of Finance 79, 3859–3891. https://doi.org/10.1111/jofi.13395

Ehsani, Sina, and Juhani T. Linnainmaa, 2022, Factor momentum and the momentum factor, Journal of Finance 77, 1877–1919. https://doi.org/10.1111/jofi.13131

Fama, Eugene F., and Kenneth R. French, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial Economics 33, 3–56. https://doi.org/10.1016/0304-405X(93)90023-5

Fleming, Jeff, Chris Kirby, and Barbara Ostdiek, 2001, The economic value of volatility timing, Journal of Finance 56, 329–352. https://doi.org/10.1111/0022-1082.00327

Fleming, Jeff, Chris Kirby, and Barbara Ostdiek, 2003, The economic value of volatility timing using "realized" volatility, Journal of Financial Economics 67, 473–509. https://doi.org/10.1016/S0304-405X(02)00259-3

French, Kenneth R., Data Library, Tuck School of Business, Dartmouth College, Fama/French 3 Factors (monthly), retrieved 24 October 2026. https://mba.tuck.dartmouth.edu/pages/faculty/ken.french/data_library.html

French, Kenneth R., G. William Schwert, and Robert F. Stambaugh, 1987, Expected stock returns and volatility, Journal of Financial Economics 19, 3–29. https://doi.org/10.1016/0304-405X(87)90026-2

Glosten, Lawrence R., Ravi Jagannathan, and David E. Runkle, 1993, On the relation between the expected value and the volatility of the nominal excess return on stocks, Journal of Finance 48, 1779–1801. https://doi.org/10.1111/j.1540-6261.1993.tb05128.x

Goyal, Amit, and Ivo Welch, 2008, A comprehensive look at the empirical performance of equity premium prediction, Review of Financial Studies 21, 1455–1508. https://doi.org/10.1093/rfs/hhm014

Gupta, Tarun, and Bryan Kelly, 2019, Factor momentum everywhere, Journal of Portfolio Management 45, 13–36. https://doi.org/10.3905/jpm.2019.45.3.013

Haddad, Valentin, Serhiy Kozak, and Shrihari Santosh, 2020, Factor timing, Review of Financial Studies 33, 1980–2018. https://doi.org/10.1093/rfs/hhaa017

Harvey, Campbell R., Edward Hoyle, Russell Korgaonkar, Sandy Rattray, Matthew Sargaison, and Otto Van Hemert, 2018, The impact of volatility targeting, Journal of Portfolio Management 45, 14–33. https://doi.org/10.3905/jpm.2018.45.1.014

Huang, Dashan, Jiangyuan Li, Liyao Wang, and Guofu Zhou, 2020, Time series momentum: Is it there?, Journal of Financial Economics 135, 774–794. https://doi.org/10.1016/j.jfineco.2019.08.004

Huberman, Gur, and Shmuel Kandel, 1987, Mean-variance spanning, Journal of Finance 42, 873–888. https://doi.org/10.1111/j.1540-6261.1987.tb03917.x

Jobson, J. D., and Bob M. Korkie, 1981, Performance hypothesis testing with the Sharpe and Treynor measures, Journal of Finance 36, 889–908. https://doi.org/10.1111/j.1540-6261.1981.tb04891.x

Kim, Abby Y., Yiuman Tse, and John K. Wald, 2016, Time series momentum and volatility scaling, Journal of Financial Markets 30, 103–124. https://doi.org/10.1016/j.finmar.2016.05.003

Ledoit, Olivier, and Michael Wolf, 2008, Robust performance hypothesis testing with the Sharpe ratio, Journal of Empirical Finance 15, 850–859. https://doi.org/10.1016/j.jempfin.2008.03.002

Liu, Fang, Xiaoxiao Tang, and Guofu Zhou, 2019, Volatility-managed portfolio: Does it really work?, Journal of Portfolio Management 46, 38–51. https://doi.org/10.3905/jpm.2019.1.107

Lo, Andrew W., 2002, The statistics of Sharpe ratios, Financial Analysts Journal 58, 36–52. https://doi.org/10.2469/faj.v58.n4.2453

Lochstoer, Lars A., and Tyler Muir, 2022, Volatility expectations and returns, Journal of Finance 77, 1055–1096. https://doi.org/10.1111/jofi.13120

McLean, R. David, and Jeffrey Pontiff, 2016, Does academic research destroy stock return predictability?, Journal of Finance 71, 5–32. https://doi.org/10.1111/jofi.12365

Moreira, Alan, and Tyler Muir, 2017, Volatility-managed portfolios, Journal of Finance 72, 1611–1644. https://doi.org/10.1111/jofi.12513

Moreira, Alan, and Tyler Muir, 2019, Should long-term investors time volatility?, Journal of Financial Economics 131, 507–527. https://doi.org/10.1016/j.jfineco.2018.09.011

Moskowitz, Tobias J., Yao Hua Ooi, and Lasse Heje Pedersen, 2012, Time series momentum, Journal of Financial Economics 104, 228–250. https://doi.org/10.1016/j.jfineco.2011.11.003

Newey, Whitney K., and Kenneth D. West, 1987, A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix, Econometrica 55, 703–708. https://doi.org/10.2307/1913610

Novy-Marx, Robert, and Mihail Velikov, 2016, A taxonomy of anomalies and their trading costs, Review of Financial Studies 29, 104–147. https://doi.org/10.1093/rfs/hhv063

Schwert, G. William, 1989, Why does stock market volatility change over time?, Journal of Finance 44, 1115–1153. https://doi.org/10.1111/j.1540-6261.1989.tb02647.x

Xu, Xia, 2026, Improving volatility-managed portfolios in real time, Critical Finance Review 15, 179–207. https://doi.org/10.1108/CFR-03-2023-2491

6.Review

Attempt 2Done

Automated review: Accept (novelty 3, rigor 4, clarity 4, reproducibility 5).

Accept
Novelty 3/5Rigor 4/5Clarity 4/5Reproducibility 5/5

Refereed by Claude Fable 5.1, a different model from the one that wrote the paper (Claude Opus 5.5).

The two proof corrections requested in the previous report are made: the abstract is within the journal's 100-word limit (98 words) and both it and the conclusion now state the levels at which the variance-forecast weight beats inverse-variance weighting, consistent with the record's p-values. Nothing else in the manuscript changed and the executed record is unchanged, so every reported number still matches it. All required changes from the three earlier reports have been met, within the constraints the coordinator imposed on the data. The paper is an honest, fully reproducible and well-bounded null result on volatility management measurable from monthly returns for the Fama-French market, size and value factors over 1926-2026, run against a sign-only factor-momentum rule in sample and in real time, with the daily-return test named as the decisive next step. Its contribution is modest for the venue and is described as such. Accept; scores unchanged.

Strengths

  • Both proof corrections are in and verified against the manuscript and the record. The abstract is 98 words and states the levels at which the variance-forecast weight beats inverse-variance weighting (5% for value, 10% for size, not the market); the conclusion gives the same levels. They match the record's p-values of 0.04, 0.09 and 0.32 for the three differences. The executed record (script and RESULT_JSON) is unchanged from the previous round, and the rest of the manuscript is unchanged, so every number in the paper still matches the record cell for cell.
  • The paper now carries, in full: a pre-committed design whose overall verdict is stated in its own terms (VM survives for none of the three factors; only the equality of real-time Sharpe ratios is left inconclusive); a real-time protocol in which every constant, threshold, combination weight and variance forecast uses data through month t only; Ledoit-Wolf HAC and block-bootstrap intervals, joint HAC Wald tests across the three factors, and Holm adjustments over three families of comparisons; return-unit magnitudes and a one-standard-deviation effect; a direct test of the noisy-signal explanation with properly hedged interpretation; the H1 alpha under all eight specifications with the one 5% cell bounded by the joint test, Holm, the momentum control and the real-time difference; and an honest statement of what the monthly record cannot test.
  • Venue conventions are met: first person plural; numbered tables with captions stating specification and sample and notes defining the standard errors; magnitudes in percent per year; every use of 'significant' with its level; H1-H3 named and revisited. All references are in the retrieved pool and all DOIs resolve.

Weaknesses

  • The contribution is a well-bounded null for volatility management measurable from monthly returns on three factors with gross returns; the decisive daily-return test and the momentum factor lie outside the record. The paper says so in the introduction, the data section, the limitations and the conclusion, and its title and abstract now promise nothing more.
  • The forecast weight rests on one linear forecast from the six-month window, and for size it is nearly constant, so the absence of a detectable gain there is close to a comparison of the factor with itself; disclosed in the limitations, not required for this record.

Required changes

—

Record

§Event timeline

  1. Run createdQuestion: “Do volatility-managed factor portfolios survive real-time implementation and a horse race against time-series factor momentum? A century of US market, size and value factors' language=en discipline=finance venue='Journal of Finance” · output language: English · models: agent:claude-opus-5-5
  2. Stage started · Literature reviewAttempt 1 · agent:claude-opus-5-5
  3. awaiting_agent · Literature reviewsearch response for literature_review: literature_review-search-1.response.json
  4. Resume requested
  5. Stage started · Literature reviewAttempt 2 · agent:claude-opus-5-5
  6. Search complete · Literature review41 sources · 20,128 characters of notes
  7. awaiting_agent · Literature reviewcomplete_json response for literature_review: literature_review-complete_json-1.response.json
  8. Resume requested
  9. Stage started · Literature reviewAttempt 3 · agent:claude-opus-5-5
  10. Search complete · Literature review41 sources · 20,128 characters of notes
  11. Stage completed · Literature review
  12. Stage started · IdeationAttempt 1 · agent:claude-opus-5-5
  13. awaiting_agent · Ideationcomplete_json response for ideation: ideation-complete_json-1.response.json
  14. Resume requested
  15. Stage started · IdeationAttempt 2 · agent:claude-opus-5-5
  16. Awaiting approval · Ideation
  17. Approved · IdeationIdea 1 selected
  18. Stage started · ExperimentationAttempt 1 · agent:claude-opus-5-5
  19. Dataset provisioned · Experimentationff_factors_monthly · 1,201 observations · 5 variables · sha256 b840dba55d31
  20. awaiting_agent · Experimentationcomplete response for experimentation: experimentation-complete-1.response.txt
  21. Resume requested
  22. Stage started · ExperimentationAttempt 2 · agent:claude-opus-5-5
  23. Dataset provisioned · Experimentationff_factors_monthly · 1,201 observations · 5 variables · sha256 b840dba55d31
  24. Experiment round · ExperimentationRound 1: exit 0; result parsed (0.948s)
  25. Stage completed · Experimentation
  26. Stage started · AnalysisAttempt 1 · agent:claude-opus-5-5
  27. awaiting_agent · Analysiscomplete_json response for analysis: analysis-complete_json-1.response.json
  28. Resume requested
  29. Stage started · AnalysisAttempt 2 · agent:claude-opus-5-5
  30. Stage completed · Analysis
  31. Stage started · Paper writingAttempt 1 · agent:claude-opus-5-5
  32. awaiting_agent · Paper writingcomplete response for paper_writing: paper_writing-complete-1.response.txt
  33. Resume requested
  34. Stage started · Paper writingAttempt 2 · agent:claude-opus-5-5
  35. Stage completed · Paper writing
  36. Stage started · ReviewAttempt 1 · agent:claude-opus-5-5
  37. awaiting_agent · Reviewcomplete_json response for review: review-complete_json-1.response.json
  38. Resume requested
  39. Stage started · ReviewAttempt 2 · agent:claude-opus-5-5
  40. Awaiting approval · Review
  41. Sent back for revision · ReviewRe-running from Paper writing: “The referee recommends minor revision. Make these required changes, quoted from the report; none requires new estimation, and every number must come from the existing RESULT_JSON:\n1. Retitle so that '”
  42. Stage started · Paper writingAttempt 3 · agent:claude-opus-5-5
  43. awaiting_agent · Paper writingcomplete response for paper_writing: paper_writing-complete-1.response.txt
  44. Resume requested
  45. Stage started · Paper writingAttempt 4 · agent:claude-opus-5-5
  46. Stage completed · Paper writing
  47. Stage started · ReviewAttempt 1 · agent:claude-opus-5-5
  48. awaiting_agent · Reviewcomplete_json response for review: review-complete_json-1.response.json
  49. Resume requested
  50. Stage started · ReviewAttempt 2 · agent:claude-opus-5-5
  51. Awaiting approval · Review
  52. Approved · Review
  53. Run completereport=/Users/haodingresearch/Desktop/auto-research/runs/studies-opus55/finance-en/runs/963f889ffb68/REPORT.md
  54. Reopened after sign-off · ReviewReopened after sign-off; re-running from Review: “Independent review: the paper is re-refereed by an independent referee (Claude Fable 5.1) that took no part in writing it.”
  55. Stage started · ReviewAttempt 3 · agent:claude-fable-5-1
  56. awaiting_agent · Reviewcomplete_json response for review: review-complete_json-1.response.json
  57. Resume requested
  58. Stage started · ReviewAttempt 4 · agent:claude-fable-5-1
  59. Awaiting approval · Review
  60. Sent back for revision · ReviewRe-running from Experimentation: “Independent referee (Claude Fable 5.1): major_revision. Required changes: (1) [analysis] Build the Moreira-Muir signal from the daily Fama-French factor file (same library, July 1926 onward): previous”
  61. Stage started · ExperimentationAttempt 3 · agent:claude-opus-5-5
  62. Dataset provisioned · Experimentationff_factors_monthly · 1,201 observations · 5 variables · sha256 b840dba55d31
  63. awaiting_agent · Experimentationcomplete response for experimentation: experimentation-complete-1.response.txt
  64. Resume requested
  65. Stage started · ExperimentationAttempt 4 · agent:claude-opus-5-5
  66. Dataset provisioned · Experimentationff_factors_monthly · 1,201 observations · 5 variables · sha256 b840dba55d31
  67. Experiment round · ExperimentationRound 1: exit 0; result parsed (0.982s)
  68. Stage completed · Experimentation
  69. Stage started · AnalysisAttempt 1 · agent:claude-opus-5-5
  70. awaiting_agent · Analysiscomplete_json response for analysis: analysis-complete_json-1.response.json
  71. Resume requested
  72. Stage started · AnalysisAttempt 2 · agent:claude-opus-5-5
  73. Stage completed · Analysis
  74. Stage started · Paper writingAttempt 5 · agent:claude-opus-5-5
  75. awaiting_agent · Paper writingcomplete response for paper_writing: paper_writing-complete-1.response.txt
  76. Resume requested
  77. Stage started · Paper writingAttempt 6 · agent:claude-opus-5-5
  78. Stage completed · Paper writing
  79. Stage started · ReviewAttempt 1 · agent:claude-fable-5-1
  80. awaiting_agent · Reviewcomplete_json response for review: review-complete_json-1.response.json
  81. Resume requested
  82. Stage started · ReviewAttempt 2 · agent:claude-fable-5-1
  83. Awaiting approval · Review
  84. Sent back for revision · ReviewRe-running from Experimentation: “Independent referee (Claude Fable 5.1), round 2: minor_revision. Required changes: (1) [analysis] Add one row to Table VII Panel A: a managed factor with weight proportional to 1/(a_t + b_t RV_t), whe”
  85. Stage started · ExperimentationAttempt 5 · agent:claude-opus-5-5
  86. Dataset provisioned · Experimentationff_factors_monthly · 1,201 observations · 5 variables · sha256 b840dba55d31
  87. awaiting_agent · Experimentationcomplete response for experimentation: experimentation-complete-1.response.txt
  88. Resume requested
  89. Stage started · ExperimentationAttempt 6 · agent:claude-opus-5-5
  90. Dataset provisioned · Experimentationff_factors_monthly · 1,201 observations · 5 variables · sha256 b840dba55d31
  91. Experiment round · ExperimentationRound 1: exit 0; result parsed (1.15s)
  92. Stage completed · Experimentation
  93. Stage started · AnalysisAttempt 1 · agent:claude-opus-5-5
  94. awaiting_agent · Analysiscomplete_json response for analysis: analysis-complete_json-1.response.json
  95. Resume requested
  96. Stage started · AnalysisAttempt 2 · agent:claude-opus-5-5
  97. Stage completed · Analysis
  98. Stage started · Paper writingAttempt 7 · agent:claude-opus-5-5
  99. awaiting_agent · Paper writingcomplete response for paper_writing: paper_writing-complete-1.response.txt
  100. Resume requested
  101. Stage started · Paper writingAttempt 8 · agent:claude-opus-5-5
  102. Stage completed · Paper writing
  103. Stage started · ReviewAttempt 1 · agent:claude-fable-5-1
  104. awaiting_agent · Reviewcomplete_json response for review: review-complete_json-1.response.json
  105. Resume requested
  106. Stage started · ReviewAttempt 2 · agent:claude-fable-5-1
  107. Awaiting approval · Review
  108. Sent back for revision · ReviewRe-running from Analysis: “Independent referee (Claude Fable 5.1), round 3: accept, with proof corrections: (1) [text] Cut the abstract to 100 words or fewer (it is 117): for example drop 'across eight specifications' and the p”
  109. Stage started · AnalysisAttempt 3 · agent:claude-opus-5-5
  110. awaiting_agent · Analysiscomplete_json response for analysis: analysis-complete_json-1.response.json
  111. Resume requested
  112. Stage started · AnalysisAttempt 4 · agent:claude-opus-5-5
  113. Stage completed · Analysis
  114. Stage started · Paper writingAttempt 9 · agent:claude-opus-5-5
  115. awaiting_agent · Paper writingcomplete response for paper_writing: paper_writing-complete-1.response.txt
  116. Resume requested
  117. Stage started · Paper writingAttempt 10 · agent:claude-opus-5-5
  118. Stage completed · Paper writing
  119. Stage started · ReviewAttempt 1 · agent:claude-fable-5-1
  120. awaiting_agent · Reviewcomplete_json response for review: review-complete_json-1.response.json
  121. Resume requested
  122. Stage started · ReviewAttempt 2 · agent:claude-fable-5-1
  123. Awaiting approval · Review
  124. Approved · Review
  125. Run completereport=/Users/haodingresearch/Desktop/auto-research/runs/studies-opus55/finance-en/runs/963f889ffb68/REPORT.md